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/sched.h>
21 #include <linux/writeback.h>
22 #include <linux/pagemap.h>
25 #include "transaction.h"
27 #include "ref-cache.h"
30 static int total_trans = 0;
31 extern struct kmem_cache *btrfs_trans_handle_cachep;
32 extern struct kmem_cache *btrfs_transaction_cachep;
34 #define BTRFS_ROOT_TRANS_TAG 0
36 static noinline void put_transaction(struct btrfs_transaction *transaction)
38 WARN_ON(transaction->use_count == 0);
39 transaction->use_count--;
40 if (transaction->use_count == 0) {
41 WARN_ON(total_trans == 0);
43 list_del_init(&transaction->list);
44 memset(transaction, 0, sizeof(*transaction));
45 kmem_cache_free(btrfs_transaction_cachep, transaction);
50 * either allocate a new transaction or hop into the existing one
52 static noinline int join_transaction(struct btrfs_root *root)
54 struct btrfs_transaction *cur_trans;
55 cur_trans = root->fs_info->running_transaction;
57 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
61 root->fs_info->generation++;
62 root->fs_info->last_alloc = 0;
63 root->fs_info->last_data_alloc = 0;
64 root->fs_info->last_log_alloc = 0;
65 cur_trans->num_writers = 1;
66 cur_trans->num_joined = 0;
67 cur_trans->transid = root->fs_info->generation;
68 init_waitqueue_head(&cur_trans->writer_wait);
69 init_waitqueue_head(&cur_trans->commit_wait);
70 cur_trans->in_commit = 0;
71 cur_trans->blocked = 0;
72 cur_trans->use_count = 1;
73 cur_trans->commit_done = 0;
74 cur_trans->start_time = get_seconds();
75 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
76 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
77 extent_io_tree_init(&cur_trans->dirty_pages,
78 root->fs_info->btree_inode->i_mapping,
80 spin_lock(&root->fs_info->new_trans_lock);
81 root->fs_info->running_transaction = cur_trans;
82 spin_unlock(&root->fs_info->new_trans_lock);
84 cur_trans->num_writers++;
85 cur_trans->num_joined++;
92 * this does all the record keeping required to make sure that a
93 * reference counted root is properly recorded in a given transaction.
94 * This is required to make sure the old root from before we joined the transaction
95 * is deleted when the transaction commits
97 noinline int btrfs_record_root_in_trans(struct btrfs_root *root)
99 struct btrfs_dirty_root *dirty;
100 u64 running_trans_id = root->fs_info->running_transaction->transid;
101 if (root->ref_cows && root->last_trans < running_trans_id) {
102 WARN_ON(root == root->fs_info->extent_root);
103 if (root->root_item.refs != 0) {
104 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
105 (unsigned long)root->root_key.objectid,
106 BTRFS_ROOT_TRANS_TAG);
108 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
110 dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
111 BUG_ON(!dirty->root);
112 dirty->latest_root = root;
113 INIT_LIST_HEAD(&dirty->list);
115 root->commit_root = btrfs_root_node(root);
117 memcpy(dirty->root, root, sizeof(*root));
118 spin_lock_init(&dirty->root->node_lock);
119 spin_lock_init(&dirty->root->list_lock);
120 mutex_init(&dirty->root->objectid_mutex);
121 mutex_init(&dirty->root->log_mutex);
122 INIT_LIST_HEAD(&dirty->root->dead_list);
123 dirty->root->node = root->commit_root;
124 dirty->root->commit_root = NULL;
126 spin_lock(&root->list_lock);
127 list_add(&dirty->root->dead_list, &root->dead_list);
128 spin_unlock(&root->list_lock);
130 root->dirty_root = dirty;
134 root->last_trans = running_trans_id;
139 /* wait for commit against the current transaction to become unblocked
140 * when this is done, it is safe to start a new transaction, but the current
141 * transaction might not be fully on disk.
143 static void wait_current_trans(struct btrfs_root *root)
145 struct btrfs_transaction *cur_trans;
147 cur_trans = root->fs_info->running_transaction;
148 if (cur_trans && cur_trans->blocked) {
150 cur_trans->use_count++;
152 prepare_to_wait(&root->fs_info->transaction_wait, &wait,
153 TASK_UNINTERRUPTIBLE);
154 if (cur_trans->blocked) {
155 mutex_unlock(&root->fs_info->trans_mutex);
157 mutex_lock(&root->fs_info->trans_mutex);
158 finish_wait(&root->fs_info->transaction_wait,
161 finish_wait(&root->fs_info->transaction_wait,
166 put_transaction(cur_trans);
170 static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
171 int num_blocks, int wait)
173 struct btrfs_trans_handle *h =
174 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
177 mutex_lock(&root->fs_info->trans_mutex);
178 if (!root->fs_info->log_root_recovering &&
179 ((wait == 1 && !root->fs_info->open_ioctl_trans) || wait == 2))
180 wait_current_trans(root);
181 ret = join_transaction(root);
184 btrfs_record_root_in_trans(root);
185 h->transid = root->fs_info->running_transaction->transid;
186 h->transaction = root->fs_info->running_transaction;
187 h->blocks_reserved = num_blocks;
189 h->block_group = NULL;
190 h->alloc_exclude_nr = 0;
191 h->alloc_exclude_start = 0;
192 root->fs_info->running_transaction->use_count++;
193 mutex_unlock(&root->fs_info->trans_mutex);
197 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
200 return start_transaction(root, num_blocks, 1);
202 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
205 return start_transaction(root, num_blocks, 0);
208 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
211 return start_transaction(r, num_blocks, 2);
214 /* wait for a transaction commit to be fully complete */
215 static noinline int wait_for_commit(struct btrfs_root *root,
216 struct btrfs_transaction *commit)
219 mutex_lock(&root->fs_info->trans_mutex);
220 while(!commit->commit_done) {
221 prepare_to_wait(&commit->commit_wait, &wait,
222 TASK_UNINTERRUPTIBLE);
223 if (commit->commit_done)
225 mutex_unlock(&root->fs_info->trans_mutex);
227 mutex_lock(&root->fs_info->trans_mutex);
229 mutex_unlock(&root->fs_info->trans_mutex);
230 finish_wait(&commit->commit_wait, &wait);
235 * rate limit against the drop_snapshot code. This helps to slow down new operations
236 * if the drop_snapshot code isn't able to keep up.
238 static void throttle_on_drops(struct btrfs_root *root)
240 struct btrfs_fs_info *info = root->fs_info;
241 int harder_count = 0;
244 if (atomic_read(&info->throttles)) {
247 thr = atomic_read(&info->throttle_gen);
250 prepare_to_wait(&info->transaction_throttle,
251 &wait, TASK_UNINTERRUPTIBLE);
252 if (!atomic_read(&info->throttles)) {
253 finish_wait(&info->transaction_throttle, &wait);
257 finish_wait(&info->transaction_throttle, &wait);
258 } while (thr == atomic_read(&info->throttle_gen));
261 if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 &&
265 if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 &&
269 if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 &&
275 void btrfs_throttle(struct btrfs_root *root)
277 mutex_lock(&root->fs_info->trans_mutex);
278 if (!root->fs_info->open_ioctl_trans)
279 wait_current_trans(root);
280 mutex_unlock(&root->fs_info->trans_mutex);
282 throttle_on_drops(root);
285 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
286 struct btrfs_root *root, int throttle)
288 struct btrfs_transaction *cur_trans;
289 struct btrfs_fs_info *info = root->fs_info;
291 mutex_lock(&info->trans_mutex);
292 cur_trans = info->running_transaction;
293 WARN_ON(cur_trans != trans->transaction);
294 WARN_ON(cur_trans->num_writers < 1);
295 cur_trans->num_writers--;
297 if (waitqueue_active(&cur_trans->writer_wait))
298 wake_up(&cur_trans->writer_wait);
299 put_transaction(cur_trans);
300 mutex_unlock(&info->trans_mutex);
301 memset(trans, 0, sizeof(*trans));
302 kmem_cache_free(btrfs_trans_handle_cachep, trans);
305 throttle_on_drops(root);
310 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
311 struct btrfs_root *root)
313 return __btrfs_end_transaction(trans, root, 0);
316 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
317 struct btrfs_root *root)
319 return __btrfs_end_transaction(trans, root, 1);
323 * when btree blocks are allocated, they have some corresponding bits set for
324 * them in one of two extent_io trees. This is used to make sure all of
325 * those extents are on disk for transaction or log commit
327 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
328 struct extent_io_tree *dirty_pages)
334 struct inode *btree_inode = root->fs_info->btree_inode;
340 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
344 while(start <= end) {
347 index = start >> PAGE_CACHE_SHIFT;
348 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
349 page = find_get_page(btree_inode->i_mapping, index);
353 btree_lock_page_hook(page);
354 if (!page->mapping) {
356 page_cache_release(page);
360 if (PageWriteback(page)) {
362 wait_on_page_writeback(page);
365 page_cache_release(page);
369 err = write_one_page(page, 0);
372 page_cache_release(page);
376 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
381 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
382 while(start <= end) {
383 index = start >> PAGE_CACHE_SHIFT;
384 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
385 page = find_get_page(btree_inode->i_mapping, index);
388 if (PageDirty(page)) {
389 btree_lock_page_hook(page);
390 wait_on_page_writeback(page);
391 err = write_one_page(page, 0);
395 wait_on_page_writeback(page);
396 page_cache_release(page);
405 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
406 struct btrfs_root *root)
408 if (!trans || !trans->transaction) {
409 struct inode *btree_inode;
410 btree_inode = root->fs_info->btree_inode;
411 return filemap_write_and_wait(btree_inode->i_mapping);
413 return btrfs_write_and_wait_marked_extents(root,
414 &trans->transaction->dirty_pages);
418 * this is used to update the root pointer in the tree of tree roots.
420 * But, in the case of the extent allocation tree, updating the root
421 * pointer may allocate blocks which may change the root of the extent
424 * So, this loops and repeats and makes sure the cowonly root didn't
425 * change while the root pointer was being updated in the metadata.
427 static int update_cowonly_root(struct btrfs_trans_handle *trans,
428 struct btrfs_root *root)
432 struct btrfs_root *tree_root = root->fs_info->tree_root;
434 btrfs_write_dirty_block_groups(trans, root);
436 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
437 if (old_root_bytenr == root->node->start)
439 btrfs_set_root_bytenr(&root->root_item,
441 btrfs_set_root_level(&root->root_item,
442 btrfs_header_level(root->node));
443 ret = btrfs_update_root(trans, tree_root,
447 btrfs_write_dirty_block_groups(trans, root);
453 * update all the cowonly tree roots on disk
455 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
456 struct btrfs_root *root)
458 struct btrfs_fs_info *fs_info = root->fs_info;
459 struct list_head *next;
461 while(!list_empty(&fs_info->dirty_cowonly_roots)) {
462 next = fs_info->dirty_cowonly_roots.next;
464 root = list_entry(next, struct btrfs_root, dirty_list);
465 update_cowonly_root(trans, root);
471 * dead roots are old snapshots that need to be deleted. This allocates
472 * a dirty root struct and adds it into the list of dead roots that need to
475 int btrfs_add_dead_root(struct btrfs_root *root, struct btrfs_root *latest)
477 struct btrfs_dirty_root *dirty;
479 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
483 dirty->latest_root = latest;
485 mutex_lock(&root->fs_info->trans_mutex);
486 list_add(&dirty->list, &latest->fs_info->dead_roots);
487 mutex_unlock(&root->fs_info->trans_mutex);
492 * at transaction commit time we need to schedule the old roots for
493 * deletion via btrfs_drop_snapshot. This runs through all the
494 * reference counted roots that were modified in the current
495 * transaction and puts them into the drop list
497 static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
498 struct radix_tree_root *radix,
499 struct list_head *list)
501 struct btrfs_dirty_root *dirty;
502 struct btrfs_root *gang[8];
503 struct btrfs_root *root;
510 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
512 BTRFS_ROOT_TRANS_TAG);
515 for (i = 0; i < ret; i++) {
517 radix_tree_tag_clear(radix,
518 (unsigned long)root->root_key.objectid,
519 BTRFS_ROOT_TRANS_TAG);
521 BUG_ON(!root->ref_tree);
522 dirty = root->dirty_root;
524 btrfs_free_log(trans, root);
525 btrfs_free_reloc_root(root);
527 if (root->commit_root == root->node) {
528 WARN_ON(root->node->start !=
529 btrfs_root_bytenr(&root->root_item));
531 free_extent_buffer(root->commit_root);
532 root->commit_root = NULL;
533 root->dirty_root = NULL;
535 spin_lock(&root->list_lock);
536 list_del_init(&dirty->root->dead_list);
537 spin_unlock(&root->list_lock);
542 /* make sure to update the root on disk
543 * so we get any updates to the block used
546 err = btrfs_update_root(trans,
547 root->fs_info->tree_root,
553 memset(&root->root_item.drop_progress, 0,
554 sizeof(struct btrfs_disk_key));
555 root->root_item.drop_level = 0;
556 root->commit_root = NULL;
557 root->dirty_root = NULL;
558 root->root_key.offset = root->fs_info->generation;
559 btrfs_set_root_bytenr(&root->root_item,
561 btrfs_set_root_level(&root->root_item,
562 btrfs_header_level(root->node));
563 err = btrfs_insert_root(trans, root->fs_info->tree_root,
569 refs = btrfs_root_refs(&dirty->root->root_item);
570 btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
571 err = btrfs_update_root(trans, root->fs_info->tree_root,
572 &dirty->root->root_key,
573 &dirty->root->root_item);
577 list_add(&dirty->list, list);
580 free_extent_buffer(dirty->root->node);
590 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
591 * otherwise every leaf in the btree is read and defragged.
593 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
595 struct btrfs_fs_info *info = root->fs_info;
597 struct btrfs_trans_handle *trans;
601 if (root->defrag_running)
603 trans = btrfs_start_transaction(root, 1);
605 root->defrag_running = 1;
606 ret = btrfs_defrag_leaves(trans, root, cacheonly);
607 nr = trans->blocks_used;
608 btrfs_end_transaction(trans, root);
609 btrfs_btree_balance_dirty(info->tree_root, nr);
612 trans = btrfs_start_transaction(root, 1);
613 if (root->fs_info->closing || ret != -EAGAIN)
616 root->defrag_running = 0;
618 btrfs_end_transaction(trans, root);
623 * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
626 static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
627 struct list_head *list)
629 struct btrfs_dirty_root *dirty;
630 struct btrfs_trans_handle *trans;
638 while(!list_empty(list)) {
639 struct btrfs_root *root;
641 dirty = list_entry(list->prev, struct btrfs_dirty_root, list);
642 list_del_init(&dirty->list);
644 num_bytes = btrfs_root_used(&dirty->root->root_item);
645 root = dirty->latest_root;
646 atomic_inc(&root->fs_info->throttles);
649 trans = btrfs_start_transaction(tree_root, 1);
650 mutex_lock(&root->fs_info->drop_mutex);
651 ret = btrfs_drop_snapshot(trans, dirty->root);
652 if (ret != -EAGAIN) {
655 mutex_unlock(&root->fs_info->drop_mutex);
657 err = btrfs_update_root(trans,
659 &dirty->root->root_key,
660 &dirty->root->root_item);
663 nr = trans->blocks_used;
664 ret = btrfs_end_transaction(trans, tree_root);
667 btrfs_btree_balance_dirty(tree_root, nr);
671 atomic_dec(&root->fs_info->throttles);
672 wake_up(&root->fs_info->transaction_throttle);
674 mutex_lock(&root->fs_info->alloc_mutex);
675 num_bytes -= btrfs_root_used(&dirty->root->root_item);
676 bytes_used = btrfs_root_used(&root->root_item);
678 btrfs_record_root_in_trans(root);
679 btrfs_set_root_used(&root->root_item,
680 bytes_used - num_bytes);
682 mutex_unlock(&root->fs_info->alloc_mutex);
684 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
689 mutex_unlock(&root->fs_info->drop_mutex);
691 spin_lock(&root->list_lock);
692 list_del_init(&dirty->root->dead_list);
693 if (!list_empty(&root->dead_list)) {
694 struct btrfs_root *oldest;
695 oldest = list_entry(root->dead_list.prev,
696 struct btrfs_root, dead_list);
697 max_useless = oldest->root_key.offset - 1;
699 max_useless = root->root_key.offset - 1;
701 spin_unlock(&root->list_lock);
703 nr = trans->blocks_used;
704 ret = btrfs_end_transaction(trans, tree_root);
707 ret = btrfs_remove_leaf_refs(root, max_useless, 0);
710 free_extent_buffer(dirty->root->node);
714 btrfs_btree_balance_dirty(tree_root, nr);
721 * new snapshots need to be created at a very specific time in the
722 * transaction commit. This does the actual creation
724 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
725 struct btrfs_fs_info *fs_info,
726 struct btrfs_pending_snapshot *pending)
728 struct btrfs_key key;
729 struct btrfs_root_item *new_root_item;
730 struct btrfs_root *tree_root = fs_info->tree_root;
731 struct btrfs_root *root = pending->root;
732 struct extent_buffer *tmp;
733 struct extent_buffer *old;
738 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
739 if (!new_root_item) {
743 ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
747 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
749 key.objectid = objectid;
750 key.offset = trans->transid;
751 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
753 old = btrfs_lock_root_node(root);
754 btrfs_cow_block(trans, root, old, NULL, 0, &old, 0);
756 btrfs_copy_root(trans, root, old, &tmp, objectid);
757 btrfs_tree_unlock(old);
758 free_extent_buffer(old);
760 btrfs_set_root_bytenr(new_root_item, tmp->start);
761 btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
762 ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
764 btrfs_tree_unlock(tmp);
765 free_extent_buffer(tmp);
770 * insert the directory item
772 key.offset = (u64)-1;
773 namelen = strlen(pending->name);
774 ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
775 pending->name, namelen,
776 root->fs_info->sb->s_root->d_inode->i_ino,
777 &key, BTRFS_FT_DIR, 0);
782 ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
783 pending->name, strlen(pending->name), objectid,
784 root->fs_info->sb->s_root->d_inode->i_ino, 0);
786 /* Invalidate existing dcache entry for new snapshot. */
787 btrfs_invalidate_dcache_root(root, pending->name, namelen);
790 kfree(new_root_item);
795 * create all the snapshots we've scheduled for creation
797 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
798 struct btrfs_fs_info *fs_info)
800 struct btrfs_pending_snapshot *pending;
801 struct list_head *head = &trans->transaction->pending_snapshots;
804 while(!list_empty(head)) {
805 pending = list_entry(head->next,
806 struct btrfs_pending_snapshot, list);
807 ret = create_pending_snapshot(trans, fs_info, pending);
809 list_del(&pending->list);
810 kfree(pending->name);
816 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
817 struct btrfs_root *root)
819 unsigned long joined = 0;
820 unsigned long timeout = 1;
821 struct btrfs_transaction *cur_trans;
822 struct btrfs_transaction *prev_trans = NULL;
823 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
824 struct list_head dirty_fs_roots;
825 struct extent_io_tree *pinned_copy;
829 INIT_LIST_HEAD(&dirty_fs_roots);
830 mutex_lock(&root->fs_info->trans_mutex);
831 if (trans->transaction->in_commit) {
832 cur_trans = trans->transaction;
833 trans->transaction->use_count++;
834 mutex_unlock(&root->fs_info->trans_mutex);
835 btrfs_end_transaction(trans, root);
837 ret = wait_for_commit(root, cur_trans);
840 mutex_lock(&root->fs_info->trans_mutex);
841 put_transaction(cur_trans);
842 mutex_unlock(&root->fs_info->trans_mutex);
847 pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
851 extent_io_tree_init(pinned_copy,
852 root->fs_info->btree_inode->i_mapping, GFP_NOFS);
854 trans->transaction->in_commit = 1;
855 trans->transaction->blocked = 1;
856 cur_trans = trans->transaction;
857 if (cur_trans->list.prev != &root->fs_info->trans_list) {
858 prev_trans = list_entry(cur_trans->list.prev,
859 struct btrfs_transaction, list);
860 if (!prev_trans->commit_done) {
861 prev_trans->use_count++;
862 mutex_unlock(&root->fs_info->trans_mutex);
864 wait_for_commit(root, prev_trans);
866 mutex_lock(&root->fs_info->trans_mutex);
867 put_transaction(prev_trans);
872 int snap_pending = 0;
873 joined = cur_trans->num_joined;
874 if (!list_empty(&trans->transaction->pending_snapshots))
877 WARN_ON(cur_trans != trans->transaction);
878 prepare_to_wait(&cur_trans->writer_wait, &wait,
879 TASK_UNINTERRUPTIBLE);
881 if (cur_trans->num_writers > 1)
882 timeout = MAX_SCHEDULE_TIMEOUT;
886 mutex_unlock(&root->fs_info->trans_mutex);
889 ret = btrfs_wait_ordered_extents(root, 1);
893 schedule_timeout(timeout);
895 mutex_lock(&root->fs_info->trans_mutex);
896 finish_wait(&cur_trans->writer_wait, &wait);
897 } while (cur_trans->num_writers > 1 ||
898 (cur_trans->num_joined != joined));
900 ret = create_pending_snapshots(trans, root->fs_info);
903 WARN_ON(cur_trans != trans->transaction);
905 /* btrfs_commit_tree_roots is responsible for getting the
906 * various roots consistent with each other. Every pointer
907 * in the tree of tree roots has to point to the most up to date
908 * root for every subvolume and other tree. So, we have to keep
909 * the tree logging code from jumping in and changing any
912 * At this point in the commit, there can't be any tree-log
913 * writers, but a little lower down we drop the trans mutex
914 * and let new people in. By holding the tree_log_mutex
915 * from now until after the super is written, we avoid races
916 * with the tree-log code.
918 mutex_lock(&root->fs_info->tree_log_mutex);
920 * keep tree reloc code from adding new reloc trees
922 mutex_lock(&root->fs_info->tree_reloc_mutex);
925 ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
929 /* add_dirty_roots gets rid of all the tree log roots, it is now
930 * safe to free the root of tree log roots
932 btrfs_free_log_root_tree(trans, root->fs_info);
934 btrfs_free_reloc_mappings(root);
936 ret = btrfs_commit_tree_roots(trans, root);
939 cur_trans = root->fs_info->running_transaction;
940 spin_lock(&root->fs_info->new_trans_lock);
941 root->fs_info->running_transaction = NULL;
942 spin_unlock(&root->fs_info->new_trans_lock);
943 btrfs_set_super_generation(&root->fs_info->super_copy,
945 btrfs_set_super_root(&root->fs_info->super_copy,
946 root->fs_info->tree_root->node->start);
947 btrfs_set_super_root_level(&root->fs_info->super_copy,
948 btrfs_header_level(root->fs_info->tree_root->node));
950 btrfs_set_super_chunk_root(&root->fs_info->super_copy,
951 chunk_root->node->start);
952 btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
953 btrfs_header_level(chunk_root->node));
955 if (!root->fs_info->log_root_recovering) {
956 btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
957 btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
960 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
961 sizeof(root->fs_info->super_copy));
963 btrfs_copy_pinned(root, pinned_copy);
965 trans->transaction->blocked = 0;
966 wake_up(&root->fs_info->transaction_throttle);
967 wake_up(&root->fs_info->transaction_wait);
969 mutex_unlock(&root->fs_info->trans_mutex);
970 ret = btrfs_write_and_wait_transaction(trans, root);
972 write_ctree_super(trans, root);
975 * the super is written, we can safely allow the tree-loggers
976 * to go about their business
978 mutex_unlock(&root->fs_info->tree_log_mutex);
980 btrfs_finish_extent_commit(trans, root, pinned_copy);
983 btrfs_drop_dead_reloc_roots(root);
984 mutex_unlock(&root->fs_info->tree_reloc_mutex);
986 mutex_lock(&root->fs_info->trans_mutex);
988 cur_trans->commit_done = 1;
989 root->fs_info->last_trans_committed = cur_trans->transid;
990 wake_up(&cur_trans->commit_wait);
991 put_transaction(cur_trans);
992 put_transaction(cur_trans);
994 list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
995 if (root->fs_info->closing)
996 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
998 mutex_unlock(&root->fs_info->trans_mutex);
999 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1001 if (root->fs_info->closing) {
1002 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
1008 * interface function to delete all the snapshots we have scheduled for deletion
1010 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1012 struct list_head dirty_roots;
1013 INIT_LIST_HEAD(&dirty_roots);
1015 mutex_lock(&root->fs_info->trans_mutex);
1016 list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
1017 mutex_unlock(&root->fs_info->trans_mutex);
1019 if (!list_empty(&dirty_roots)) {
1020 drop_dirty_roots(root, &dirty_roots);