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 {
168 static int may_wait_transaction(struct btrfs_root *root, int type)
170 if (!root->fs_info->log_root_recovering &&
171 ((type == TRANS_START && !root->fs_info->open_ioctl_trans) ||
172 type == TRANS_USERSPACE))
177 static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
178 u64 num_items, int type)
180 struct btrfs_trans_handle *h;
181 struct btrfs_transaction *cur_trans;
184 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
186 return ERR_PTR(-ENOMEM);
188 mutex_lock(&root->fs_info->trans_mutex);
189 if (may_wait_transaction(root, type))
190 wait_current_trans(root);
192 ret = join_transaction(root);
195 cur_trans = root->fs_info->running_transaction;
196 cur_trans->use_count++;
197 mutex_unlock(&root->fs_info->trans_mutex);
199 h->transid = cur_trans->transid;
200 h->transaction = cur_trans;
203 h->bytes_reserved = 0;
204 h->delayed_ref_updates = 0;
208 if (cur_trans->blocked && may_wait_transaction(root, type)) {
209 btrfs_commit_transaction(h, root);
214 ret = btrfs_trans_reserve_metadata(h, root, num_items);
215 if (ret == -EAGAIN) {
216 btrfs_commit_transaction(h, root);
220 btrfs_end_transaction(h, root);
225 mutex_lock(&root->fs_info->trans_mutex);
226 record_root_in_trans(h, root);
227 mutex_unlock(&root->fs_info->trans_mutex);
229 if (!current->journal_info && type != TRANS_USERSPACE)
230 current->journal_info = h;
234 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
237 return start_transaction(root, num_items, TRANS_START);
239 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
242 return start_transaction(root, 0, TRANS_JOIN);
245 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
248 return start_transaction(r, 0, TRANS_USERSPACE);
251 /* wait for a transaction commit to be fully complete */
252 static noinline int wait_for_commit(struct btrfs_root *root,
253 struct btrfs_transaction *commit)
256 mutex_lock(&root->fs_info->trans_mutex);
257 while (!commit->commit_done) {
258 prepare_to_wait(&commit->commit_wait, &wait,
259 TASK_UNINTERRUPTIBLE);
260 if (commit->commit_done)
262 mutex_unlock(&root->fs_info->trans_mutex);
264 mutex_lock(&root->fs_info->trans_mutex);
266 mutex_unlock(&root->fs_info->trans_mutex);
267 finish_wait(&commit->commit_wait, &wait);
273 * rate limit against the drop_snapshot code. This helps to slow down new
274 * operations if the drop_snapshot code isn't able to keep up.
276 static void throttle_on_drops(struct btrfs_root *root)
278 struct btrfs_fs_info *info = root->fs_info;
279 int harder_count = 0;
282 if (atomic_read(&info->throttles)) {
285 thr = atomic_read(&info->throttle_gen);
288 prepare_to_wait(&info->transaction_throttle,
289 &wait, TASK_UNINTERRUPTIBLE);
290 if (!atomic_read(&info->throttles)) {
291 finish_wait(&info->transaction_throttle, &wait);
295 finish_wait(&info->transaction_throttle, &wait);
296 } while (thr == atomic_read(&info->throttle_gen));
299 if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 &&
303 if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 &&
307 if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 &&
314 void btrfs_throttle(struct btrfs_root *root)
316 mutex_lock(&root->fs_info->trans_mutex);
317 if (!root->fs_info->open_ioctl_trans)
318 wait_current_trans(root);
319 mutex_unlock(&root->fs_info->trans_mutex);
322 static int should_end_transaction(struct btrfs_trans_handle *trans,
323 struct btrfs_root *root)
326 ret = btrfs_block_rsv_check(trans, root,
327 &root->fs_info->global_block_rsv, 0, 5);
331 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
332 struct btrfs_root *root)
334 struct btrfs_transaction *cur_trans = trans->transaction;
337 if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
340 updates = trans->delayed_ref_updates;
341 trans->delayed_ref_updates = 0;
343 btrfs_run_delayed_refs(trans, root, updates);
345 return should_end_transaction(trans, root);
348 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
349 struct btrfs_root *root, int throttle)
351 struct btrfs_transaction *cur_trans = trans->transaction;
352 struct btrfs_fs_info *info = root->fs_info;
356 unsigned long cur = trans->delayed_ref_updates;
357 trans->delayed_ref_updates = 0;
359 trans->transaction->delayed_refs.num_heads_ready > 64) {
360 trans->delayed_ref_updates = 0;
363 * do a full flush if the transaction is trying
366 if (trans->transaction->delayed_refs.flushing)
368 btrfs_run_delayed_refs(trans, root, cur);
375 btrfs_trans_release_metadata(trans, root);
377 if (!root->fs_info->open_ioctl_trans &&
378 should_end_transaction(trans, root))
379 trans->transaction->blocked = 1;
381 if (cur_trans->blocked && !cur_trans->in_commit) {
383 return btrfs_commit_transaction(trans, root);
385 wake_up_process(info->transaction_kthread);
388 mutex_lock(&info->trans_mutex);
389 WARN_ON(cur_trans != info->running_transaction);
390 WARN_ON(cur_trans->num_writers < 1);
391 cur_trans->num_writers--;
393 if (waitqueue_active(&cur_trans->writer_wait))
394 wake_up(&cur_trans->writer_wait);
395 put_transaction(cur_trans);
396 mutex_unlock(&info->trans_mutex);
398 if (current->journal_info == trans)
399 current->journal_info = NULL;
400 memset(trans, 0, sizeof(*trans));
401 kmem_cache_free(btrfs_trans_handle_cachep, trans);
404 btrfs_run_delayed_iputs(root);
409 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
410 struct btrfs_root *root)
412 return __btrfs_end_transaction(trans, root, 0);
415 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
416 struct btrfs_root *root)
418 return __btrfs_end_transaction(trans, root, 1);
422 * when btree blocks are allocated, they have some corresponding bits set for
423 * them in one of two extent_io trees. This is used to make sure all of
424 * those extents are sent to disk but does not wait on them
426 int btrfs_write_marked_extents(struct btrfs_root *root,
427 struct extent_io_tree *dirty_pages, int mark)
433 struct inode *btree_inode = root->fs_info->btree_inode;
439 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
443 while (start <= end) {
446 index = start >> PAGE_CACHE_SHIFT;
447 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
448 page = find_get_page(btree_inode->i_mapping, index);
452 btree_lock_page_hook(page);
453 if (!page->mapping) {
455 page_cache_release(page);
459 if (PageWriteback(page)) {
461 wait_on_page_writeback(page);
464 page_cache_release(page);
468 err = write_one_page(page, 0);
471 page_cache_release(page);
480 * when btree blocks are allocated, they have some corresponding bits set for
481 * them in one of two extent_io trees. This is used to make sure all of
482 * those extents are on disk for transaction or log commit. We wait
483 * on all the pages and clear them from the dirty pages state tree
485 int btrfs_wait_marked_extents(struct btrfs_root *root,
486 struct extent_io_tree *dirty_pages, int mark)
492 struct inode *btree_inode = root->fs_info->btree_inode;
498 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
503 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
504 while (start <= end) {
505 index = start >> PAGE_CACHE_SHIFT;
506 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
507 page = find_get_page(btree_inode->i_mapping, index);
510 if (PageDirty(page)) {
511 btree_lock_page_hook(page);
512 wait_on_page_writeback(page);
513 err = write_one_page(page, 0);
517 wait_on_page_writeback(page);
518 page_cache_release(page);
528 * when btree blocks are allocated, they have some corresponding bits set for
529 * them in one of two extent_io trees. This is used to make sure all of
530 * those extents are on disk for transaction or log commit
532 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
533 struct extent_io_tree *dirty_pages, int mark)
538 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
539 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
543 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
544 struct btrfs_root *root)
546 if (!trans || !trans->transaction) {
547 struct inode *btree_inode;
548 btree_inode = root->fs_info->btree_inode;
549 return filemap_write_and_wait(btree_inode->i_mapping);
551 return btrfs_write_and_wait_marked_extents(root,
552 &trans->transaction->dirty_pages,
557 * this is used to update the root pointer in the tree of tree roots.
559 * But, in the case of the extent allocation tree, updating the root
560 * pointer may allocate blocks which may change the root of the extent
563 * So, this loops and repeats and makes sure the cowonly root didn't
564 * change while the root pointer was being updated in the metadata.
566 static int update_cowonly_root(struct btrfs_trans_handle *trans,
567 struct btrfs_root *root)
572 struct btrfs_root *tree_root = root->fs_info->tree_root;
574 old_root_used = btrfs_root_used(&root->root_item);
575 btrfs_write_dirty_block_groups(trans, root);
578 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
579 if (old_root_bytenr == root->node->start &&
580 old_root_used == btrfs_root_used(&root->root_item))
583 btrfs_set_root_node(&root->root_item, root->node);
584 ret = btrfs_update_root(trans, tree_root,
589 old_root_used = btrfs_root_used(&root->root_item);
590 ret = btrfs_write_dirty_block_groups(trans, root);
594 if (root != root->fs_info->extent_root)
595 switch_commit_root(root);
601 * update all the cowonly tree roots on disk
603 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
604 struct btrfs_root *root)
606 struct btrfs_fs_info *fs_info = root->fs_info;
607 struct list_head *next;
608 struct extent_buffer *eb;
611 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
614 eb = btrfs_lock_root_node(fs_info->tree_root);
615 btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
616 btrfs_tree_unlock(eb);
617 free_extent_buffer(eb);
619 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
622 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
623 next = fs_info->dirty_cowonly_roots.next;
625 root = list_entry(next, struct btrfs_root, dirty_list);
627 update_cowonly_root(trans, root);
630 down_write(&fs_info->extent_commit_sem);
631 switch_commit_root(fs_info->extent_root);
632 up_write(&fs_info->extent_commit_sem);
638 * dead roots are old snapshots that need to be deleted. This allocates
639 * a dirty root struct and adds it into the list of dead roots that need to
642 int btrfs_add_dead_root(struct btrfs_root *root)
644 mutex_lock(&root->fs_info->trans_mutex);
645 list_add(&root->root_list, &root->fs_info->dead_roots);
646 mutex_unlock(&root->fs_info->trans_mutex);
651 * update all the cowonly tree roots on disk
653 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
654 struct btrfs_root *root)
656 struct btrfs_root *gang[8];
657 struct btrfs_fs_info *fs_info = root->fs_info;
663 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
666 BTRFS_ROOT_TRANS_TAG);
669 for (i = 0; i < ret; i++) {
671 radix_tree_tag_clear(&fs_info->fs_roots_radix,
672 (unsigned long)root->root_key.objectid,
673 BTRFS_ROOT_TRANS_TAG);
675 btrfs_free_log(trans, root);
676 btrfs_update_reloc_root(trans, root);
677 btrfs_orphan_commit_root(trans, root);
679 if (root->commit_root != root->node) {
680 switch_commit_root(root);
681 btrfs_set_root_node(&root->root_item,
685 err = btrfs_update_root(trans, fs_info->tree_root,
696 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
697 * otherwise every leaf in the btree is read and defragged.
699 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
701 struct btrfs_fs_info *info = root->fs_info;
702 struct btrfs_trans_handle *trans;
706 if (xchg(&root->defrag_running, 1))
710 trans = btrfs_start_transaction(root, 0);
712 return PTR_ERR(trans);
714 ret = btrfs_defrag_leaves(trans, root, cacheonly);
716 nr = trans->blocks_used;
717 btrfs_end_transaction(trans, root);
718 btrfs_btree_balance_dirty(info->tree_root, nr);
721 if (root->fs_info->closing || ret != -EAGAIN)
724 root->defrag_running = 0;
730 * when dropping snapshots, we generate a ton of delayed refs, and it makes
731 * sense not to join the transaction while it is trying to flush the current
732 * queue of delayed refs out.
734 * This is used by the drop snapshot code only
736 static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info)
740 mutex_lock(&info->trans_mutex);
741 while (info->running_transaction &&
742 info->running_transaction->delayed_refs.flushing) {
743 prepare_to_wait(&info->transaction_wait, &wait,
744 TASK_UNINTERRUPTIBLE);
745 mutex_unlock(&info->trans_mutex);
749 mutex_lock(&info->trans_mutex);
750 finish_wait(&info->transaction_wait, &wait);
752 mutex_unlock(&info->trans_mutex);
757 * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
760 int btrfs_drop_dead_root(struct btrfs_root *root)
762 struct btrfs_trans_handle *trans;
763 struct btrfs_root *tree_root = root->fs_info->tree_root;
769 * we don't want to jump in and create a bunch of
770 * delayed refs if the transaction is starting to close
772 wait_transaction_pre_flush(tree_root->fs_info);
773 trans = btrfs_start_transaction(tree_root, 1);
776 * we've joined a transaction, make sure it isn't
779 if (trans->transaction->delayed_refs.flushing) {
780 btrfs_end_transaction(trans, tree_root);
784 ret = btrfs_drop_snapshot(trans, root);
788 ret = btrfs_update_root(trans, tree_root,
794 nr = trans->blocks_used;
795 ret = btrfs_end_transaction(trans, tree_root);
798 btrfs_btree_balance_dirty(tree_root, nr);
803 ret = btrfs_del_root(trans, tree_root, &root->root_key);
806 nr = trans->blocks_used;
807 ret = btrfs_end_transaction(trans, tree_root);
810 free_extent_buffer(root->node);
811 free_extent_buffer(root->commit_root);
814 btrfs_btree_balance_dirty(tree_root, nr);
820 * new snapshots need to be created at a very specific time in the
821 * transaction commit. This does the actual creation
823 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
824 struct btrfs_fs_info *fs_info,
825 struct btrfs_pending_snapshot *pending)
827 struct btrfs_key key;
828 struct btrfs_root_item *new_root_item;
829 struct btrfs_root *tree_root = fs_info->tree_root;
830 struct btrfs_root *root = pending->root;
831 struct btrfs_root *parent_root;
832 struct inode *parent_inode;
833 struct dentry *dentry;
834 struct extent_buffer *tmp;
835 struct extent_buffer *old;
841 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
842 if (!new_root_item) {
843 pending->error = -ENOMEM;
847 ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
849 pending->error = ret;
853 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
854 btrfs_orphan_pre_snapshot(trans, pending, &to_reserve);
856 if (to_reserve > 0) {
857 ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv,
860 pending->error = ret;
865 key.objectid = objectid;
866 key.offset = (u64)-1;
867 key.type = BTRFS_ROOT_ITEM_KEY;
869 trans->block_rsv = &pending->block_rsv;
871 dentry = pending->dentry;
872 parent_inode = dentry->d_parent->d_inode;
873 parent_root = BTRFS_I(parent_inode)->root;
874 record_root_in_trans(trans, parent_root);
877 * insert the directory item
879 ret = btrfs_set_inode_index(parent_inode, &index);
881 ret = btrfs_insert_dir_item(trans, parent_root,
882 dentry->d_name.name, dentry->d_name.len,
883 parent_inode->i_ino, &key,
884 BTRFS_FT_DIR, index);
887 btrfs_i_size_write(parent_inode, parent_inode->i_size +
888 dentry->d_name.len * 2);
889 ret = btrfs_update_inode(trans, parent_root, parent_inode);
892 record_root_in_trans(trans, root);
893 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
894 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
896 old = btrfs_lock_root_node(root);
897 btrfs_cow_block(trans, root, old, NULL, 0, &old);
898 btrfs_set_lock_blocking(old);
900 btrfs_copy_root(trans, root, old, &tmp, objectid);
901 btrfs_tree_unlock(old);
902 free_extent_buffer(old);
904 btrfs_set_root_node(new_root_item, tmp);
905 /* record when the snapshot was created in key.offset */
906 key.offset = trans->transid;
907 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
908 btrfs_tree_unlock(tmp);
909 free_extent_buffer(tmp);
913 * insert root back/forward references
915 ret = btrfs_add_root_ref(trans, tree_root, objectid,
916 parent_root->root_key.objectid,
917 parent_inode->i_ino, index,
918 dentry->d_name.name, dentry->d_name.len);
921 key.offset = (u64)-1;
922 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
923 BUG_ON(IS_ERR(pending->snap));
925 btrfs_reloc_post_snapshot(trans, pending);
926 btrfs_orphan_post_snapshot(trans, pending);
928 kfree(new_root_item);
929 btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
934 * create all the snapshots we've scheduled for creation
936 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
937 struct btrfs_fs_info *fs_info)
939 struct btrfs_pending_snapshot *pending;
940 struct list_head *head = &trans->transaction->pending_snapshots;
943 list_for_each_entry(pending, head, list) {
944 ret = create_pending_snapshot(trans, fs_info, pending);
950 static void update_super_roots(struct btrfs_root *root)
952 struct btrfs_root_item *root_item;
953 struct btrfs_super_block *super;
955 super = &root->fs_info->super_copy;
957 root_item = &root->fs_info->chunk_root->root_item;
958 super->chunk_root = root_item->bytenr;
959 super->chunk_root_generation = root_item->generation;
960 super->chunk_root_level = root_item->level;
962 root_item = &root->fs_info->tree_root->root_item;
963 super->root = root_item->bytenr;
964 super->generation = root_item->generation;
965 super->root_level = root_item->level;
968 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
971 spin_lock(&info->new_trans_lock);
972 if (info->running_transaction)
973 ret = info->running_transaction->in_commit;
974 spin_unlock(&info->new_trans_lock);
978 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
981 spin_lock(&info->new_trans_lock);
982 if (info->running_transaction)
983 ret = info->running_transaction->blocked;
984 spin_unlock(&info->new_trans_lock);
988 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
989 struct btrfs_root *root)
991 unsigned long joined = 0;
992 unsigned long timeout = 1;
993 struct btrfs_transaction *cur_trans;
994 struct btrfs_transaction *prev_trans = NULL;
998 unsigned long now = get_seconds();
999 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
1001 btrfs_run_ordered_operations(root, 0);
1003 /* make a pass through all the delayed refs we have so far
1004 * any runnings procs may add more while we are here
1006 ret = btrfs_run_delayed_refs(trans, root, 0);
1009 btrfs_trans_release_metadata(trans, root);
1011 cur_trans = trans->transaction;
1013 * set the flushing flag so procs in this transaction have to
1014 * start sending their work down.
1016 cur_trans->delayed_refs.flushing = 1;
1018 ret = btrfs_run_delayed_refs(trans, root, 0);
1021 mutex_lock(&root->fs_info->trans_mutex);
1022 if (cur_trans->in_commit) {
1023 cur_trans->use_count++;
1024 mutex_unlock(&root->fs_info->trans_mutex);
1025 btrfs_end_transaction(trans, root);
1027 ret = wait_for_commit(root, cur_trans);
1030 mutex_lock(&root->fs_info->trans_mutex);
1031 put_transaction(cur_trans);
1032 mutex_unlock(&root->fs_info->trans_mutex);
1037 trans->transaction->in_commit = 1;
1038 trans->transaction->blocked = 1;
1039 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1040 prev_trans = list_entry(cur_trans->list.prev,
1041 struct btrfs_transaction, list);
1042 if (!prev_trans->commit_done) {
1043 prev_trans->use_count++;
1044 mutex_unlock(&root->fs_info->trans_mutex);
1046 wait_for_commit(root, prev_trans);
1048 mutex_lock(&root->fs_info->trans_mutex);
1049 put_transaction(prev_trans);
1053 if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
1057 int snap_pending = 0;
1058 joined = cur_trans->num_joined;
1059 if (!list_empty(&trans->transaction->pending_snapshots))
1062 WARN_ON(cur_trans != trans->transaction);
1063 if (cur_trans->num_writers > 1)
1064 timeout = MAX_SCHEDULE_TIMEOUT;
1065 else if (should_grow)
1068 mutex_unlock(&root->fs_info->trans_mutex);
1070 if (flush_on_commit || snap_pending) {
1071 btrfs_start_delalloc_inodes(root, 1);
1072 ret = btrfs_wait_ordered_extents(root, 0, 1);
1077 * rename don't use btrfs_join_transaction, so, once we
1078 * set the transaction to blocked above, we aren't going
1079 * to get any new ordered operations. We can safely run
1080 * it here and no for sure that nothing new will be added
1083 btrfs_run_ordered_operations(root, 1);
1085 prepare_to_wait(&cur_trans->writer_wait, &wait,
1086 TASK_UNINTERRUPTIBLE);
1089 if (cur_trans->num_writers > 1 || should_grow)
1090 schedule_timeout(timeout);
1092 mutex_lock(&root->fs_info->trans_mutex);
1093 finish_wait(&cur_trans->writer_wait, &wait);
1094 } while (cur_trans->num_writers > 1 ||
1095 (should_grow && cur_trans->num_joined != joined));
1097 ret = create_pending_snapshots(trans, root->fs_info);
1100 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1103 WARN_ON(cur_trans != trans->transaction);
1105 /* btrfs_commit_tree_roots is responsible for getting the
1106 * various roots consistent with each other. Every pointer
1107 * in the tree of tree roots has to point to the most up to date
1108 * root for every subvolume and other tree. So, we have to keep
1109 * the tree logging code from jumping in and changing any
1112 * At this point in the commit, there can't be any tree-log
1113 * writers, but a little lower down we drop the trans mutex
1114 * and let new people in. By holding the tree_log_mutex
1115 * from now until after the super is written, we avoid races
1116 * with the tree-log code.
1118 mutex_lock(&root->fs_info->tree_log_mutex);
1120 ret = commit_fs_roots(trans, root);
1123 /* commit_fs_roots gets rid of all the tree log roots, it is now
1124 * safe to free the root of tree log roots
1126 btrfs_free_log_root_tree(trans, root->fs_info);
1128 ret = commit_cowonly_roots(trans, root);
1131 btrfs_prepare_extent_commit(trans, root);
1133 cur_trans = root->fs_info->running_transaction;
1134 spin_lock(&root->fs_info->new_trans_lock);
1135 root->fs_info->running_transaction = NULL;
1136 spin_unlock(&root->fs_info->new_trans_lock);
1138 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1139 root->fs_info->tree_root->node);
1140 switch_commit_root(root->fs_info->tree_root);
1142 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1143 root->fs_info->chunk_root->node);
1144 switch_commit_root(root->fs_info->chunk_root);
1146 update_super_roots(root);
1148 if (!root->fs_info->log_root_recovering) {
1149 btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
1150 btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
1153 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
1154 sizeof(root->fs_info->super_copy));
1156 trans->transaction->blocked = 0;
1158 wake_up(&root->fs_info->transaction_wait);
1160 mutex_unlock(&root->fs_info->trans_mutex);
1161 ret = btrfs_write_and_wait_transaction(trans, root);
1163 write_ctree_super(trans, root, 0);
1166 * the super is written, we can safely allow the tree-loggers
1167 * to go about their business
1169 mutex_unlock(&root->fs_info->tree_log_mutex);
1171 btrfs_finish_extent_commit(trans, root);
1173 mutex_lock(&root->fs_info->trans_mutex);
1175 cur_trans->commit_done = 1;
1177 root->fs_info->last_trans_committed = cur_trans->transid;
1179 wake_up(&cur_trans->commit_wait);
1181 put_transaction(cur_trans);
1182 put_transaction(cur_trans);
1184 mutex_unlock(&root->fs_info->trans_mutex);
1186 if (current->journal_info == trans)
1187 current->journal_info = NULL;
1189 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1191 if (current != root->fs_info->transaction_kthread)
1192 btrfs_run_delayed_iputs(root);
1198 * interface function to delete all the snapshots we have scheduled for deletion
1200 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1203 struct btrfs_fs_info *fs_info = root->fs_info;
1205 mutex_lock(&fs_info->trans_mutex);
1206 list_splice_init(&fs_info->dead_roots, &list);
1207 mutex_unlock(&fs_info->trans_mutex);
1209 while (!list_empty(&list)) {
1210 root = list_entry(list.next, struct btrfs_root, root_list);
1211 list_del(&root->root_list);
1213 if (btrfs_header_backref_rev(root->node) <
1214 BTRFS_MIXED_BACKREF_REV)
1215 btrfs_drop_snapshot(root, NULL, 0);
1217 btrfs_drop_snapshot(root, NULL, 1);