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);
284 * rate limit against the drop_snapshot code. This helps to slow down new
285 * operations if the drop_snapshot code isn't able to keep up.
287 static void throttle_on_drops(struct btrfs_root *root)
289 struct btrfs_fs_info *info = root->fs_info;
290 int harder_count = 0;
293 if (atomic_read(&info->throttles)) {
296 thr = atomic_read(&info->throttle_gen);
299 prepare_to_wait(&info->transaction_throttle,
300 &wait, TASK_UNINTERRUPTIBLE);
301 if (!atomic_read(&info->throttles)) {
302 finish_wait(&info->transaction_throttle, &wait);
306 finish_wait(&info->transaction_throttle, &wait);
307 } while (thr == atomic_read(&info->throttle_gen));
310 if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 &&
314 if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 &&
318 if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 &&
325 void btrfs_throttle(struct btrfs_root *root)
327 mutex_lock(&root->fs_info->trans_mutex);
328 if (!root->fs_info->open_ioctl_trans)
329 wait_current_trans(root);
330 mutex_unlock(&root->fs_info->trans_mutex);
333 static int should_end_transaction(struct btrfs_trans_handle *trans,
334 struct btrfs_root *root)
337 ret = btrfs_block_rsv_check(trans, root,
338 &root->fs_info->global_block_rsv, 0, 5);
342 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
343 struct btrfs_root *root)
345 struct btrfs_transaction *cur_trans = trans->transaction;
348 if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
351 updates = trans->delayed_ref_updates;
352 trans->delayed_ref_updates = 0;
354 btrfs_run_delayed_refs(trans, root, updates);
356 return should_end_transaction(trans, root);
359 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
360 struct btrfs_root *root, int throttle, int lock)
362 struct btrfs_transaction *cur_trans = trans->transaction;
363 struct btrfs_fs_info *info = root->fs_info;
367 unsigned long cur = trans->delayed_ref_updates;
368 trans->delayed_ref_updates = 0;
370 trans->transaction->delayed_refs.num_heads_ready > 64) {
371 trans->delayed_ref_updates = 0;
374 * do a full flush if the transaction is trying
377 if (trans->transaction->delayed_refs.flushing)
379 btrfs_run_delayed_refs(trans, root, cur);
386 btrfs_trans_release_metadata(trans, root);
388 if (lock && !root->fs_info->open_ioctl_trans &&
389 should_end_transaction(trans, root))
390 trans->transaction->blocked = 1;
392 if (lock && cur_trans->blocked && !cur_trans->in_commit) {
394 return btrfs_commit_transaction(trans, root);
396 wake_up_process(info->transaction_kthread);
400 mutex_lock(&info->trans_mutex);
401 WARN_ON(cur_trans != info->running_transaction);
402 WARN_ON(cur_trans->num_writers < 1);
403 cur_trans->num_writers--;
405 if (waitqueue_active(&cur_trans->writer_wait))
406 wake_up(&cur_trans->writer_wait);
407 put_transaction(cur_trans);
409 mutex_unlock(&info->trans_mutex);
411 if (current->journal_info == trans)
412 current->journal_info = NULL;
413 memset(trans, 0, sizeof(*trans));
414 kmem_cache_free(btrfs_trans_handle_cachep, trans);
417 btrfs_run_delayed_iputs(root);
422 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
423 struct btrfs_root *root)
425 return __btrfs_end_transaction(trans, root, 0, 1);
428 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
429 struct btrfs_root *root)
431 return __btrfs_end_transaction(trans, root, 1, 1);
434 int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
435 struct btrfs_root *root)
437 return __btrfs_end_transaction(trans, root, 0, 0);
441 * when btree blocks are allocated, they have some corresponding bits set for
442 * them in one of two extent_io trees. This is used to make sure all of
443 * those extents are sent to disk but does not wait on them
445 int btrfs_write_marked_extents(struct btrfs_root *root,
446 struct extent_io_tree *dirty_pages, int mark)
452 struct inode *btree_inode = root->fs_info->btree_inode;
458 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
462 while (start <= end) {
465 index = start >> PAGE_CACHE_SHIFT;
466 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
467 page = find_get_page(btree_inode->i_mapping, index);
471 btree_lock_page_hook(page);
472 if (!page->mapping) {
474 page_cache_release(page);
478 if (PageWriteback(page)) {
480 wait_on_page_writeback(page);
483 page_cache_release(page);
487 err = write_one_page(page, 0);
490 page_cache_release(page);
499 * when btree blocks are allocated, they have some corresponding bits set for
500 * them in one of two extent_io trees. This is used to make sure all of
501 * those extents are on disk for transaction or log commit. We wait
502 * on all the pages and clear them from the dirty pages state tree
504 int btrfs_wait_marked_extents(struct btrfs_root *root,
505 struct extent_io_tree *dirty_pages, int mark)
511 struct inode *btree_inode = root->fs_info->btree_inode;
517 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
522 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
523 while (start <= end) {
524 index = start >> PAGE_CACHE_SHIFT;
525 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
526 page = find_get_page(btree_inode->i_mapping, index);
529 if (PageDirty(page)) {
530 btree_lock_page_hook(page);
531 wait_on_page_writeback(page);
532 err = write_one_page(page, 0);
536 wait_on_page_writeback(page);
537 page_cache_release(page);
547 * when btree blocks are allocated, they have some corresponding bits set for
548 * them in one of two extent_io trees. This is used to make sure all of
549 * those extents are on disk for transaction or log commit
551 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
552 struct extent_io_tree *dirty_pages, int mark)
557 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
558 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
562 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
563 struct btrfs_root *root)
565 if (!trans || !trans->transaction) {
566 struct inode *btree_inode;
567 btree_inode = root->fs_info->btree_inode;
568 return filemap_write_and_wait(btree_inode->i_mapping);
570 return btrfs_write_and_wait_marked_extents(root,
571 &trans->transaction->dirty_pages,
576 * this is used to update the root pointer in the tree of tree roots.
578 * But, in the case of the extent allocation tree, updating the root
579 * pointer may allocate blocks which may change the root of the extent
582 * So, this loops and repeats and makes sure the cowonly root didn't
583 * change while the root pointer was being updated in the metadata.
585 static int update_cowonly_root(struct btrfs_trans_handle *trans,
586 struct btrfs_root *root)
591 struct btrfs_root *tree_root = root->fs_info->tree_root;
593 old_root_used = btrfs_root_used(&root->root_item);
594 btrfs_write_dirty_block_groups(trans, root);
597 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
598 if (old_root_bytenr == root->node->start &&
599 old_root_used == btrfs_root_used(&root->root_item))
602 btrfs_set_root_node(&root->root_item, root->node);
603 ret = btrfs_update_root(trans, tree_root,
608 old_root_used = btrfs_root_used(&root->root_item);
609 ret = btrfs_write_dirty_block_groups(trans, root);
613 if (root != root->fs_info->extent_root)
614 switch_commit_root(root);
620 * update all the cowonly tree roots on disk
622 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
623 struct btrfs_root *root)
625 struct btrfs_fs_info *fs_info = root->fs_info;
626 struct list_head *next;
627 struct extent_buffer *eb;
630 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
633 eb = btrfs_lock_root_node(fs_info->tree_root);
634 btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
635 btrfs_tree_unlock(eb);
636 free_extent_buffer(eb);
638 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
641 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
642 next = fs_info->dirty_cowonly_roots.next;
644 root = list_entry(next, struct btrfs_root, dirty_list);
646 update_cowonly_root(trans, root);
649 down_write(&fs_info->extent_commit_sem);
650 switch_commit_root(fs_info->extent_root);
651 up_write(&fs_info->extent_commit_sem);
657 * dead roots are old snapshots that need to be deleted. This allocates
658 * a dirty root struct and adds it into the list of dead roots that need to
661 int btrfs_add_dead_root(struct btrfs_root *root)
663 mutex_lock(&root->fs_info->trans_mutex);
664 list_add(&root->root_list, &root->fs_info->dead_roots);
665 mutex_unlock(&root->fs_info->trans_mutex);
670 * update all the cowonly tree roots on disk
672 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
673 struct btrfs_root *root)
675 struct btrfs_root *gang[8];
676 struct btrfs_fs_info *fs_info = root->fs_info;
682 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
685 BTRFS_ROOT_TRANS_TAG);
688 for (i = 0; i < ret; i++) {
690 radix_tree_tag_clear(&fs_info->fs_roots_radix,
691 (unsigned long)root->root_key.objectid,
692 BTRFS_ROOT_TRANS_TAG);
694 btrfs_free_log(trans, root);
695 btrfs_update_reloc_root(trans, root);
696 btrfs_orphan_commit_root(trans, root);
698 if (root->commit_root != root->node) {
699 switch_commit_root(root);
700 btrfs_set_root_node(&root->root_item,
704 err = btrfs_update_root(trans, fs_info->tree_root,
715 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
716 * otherwise every leaf in the btree is read and defragged.
718 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
720 struct btrfs_fs_info *info = root->fs_info;
721 struct btrfs_trans_handle *trans;
725 if (xchg(&root->defrag_running, 1))
729 trans = btrfs_start_transaction(root, 0);
731 return PTR_ERR(trans);
733 ret = btrfs_defrag_leaves(trans, root, cacheonly);
735 nr = trans->blocks_used;
736 btrfs_end_transaction(trans, root);
737 btrfs_btree_balance_dirty(info->tree_root, nr);
740 if (root->fs_info->closing || ret != -EAGAIN)
743 root->defrag_running = 0;
749 * when dropping snapshots, we generate a ton of delayed refs, and it makes
750 * sense not to join the transaction while it is trying to flush the current
751 * queue of delayed refs out.
753 * This is used by the drop snapshot code only
755 static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info)
759 mutex_lock(&info->trans_mutex);
760 while (info->running_transaction &&
761 info->running_transaction->delayed_refs.flushing) {
762 prepare_to_wait(&info->transaction_wait, &wait,
763 TASK_UNINTERRUPTIBLE);
764 mutex_unlock(&info->trans_mutex);
768 mutex_lock(&info->trans_mutex);
769 finish_wait(&info->transaction_wait, &wait);
771 mutex_unlock(&info->trans_mutex);
776 * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
779 int btrfs_drop_dead_root(struct btrfs_root *root)
781 struct btrfs_trans_handle *trans;
782 struct btrfs_root *tree_root = root->fs_info->tree_root;
788 * we don't want to jump in and create a bunch of
789 * delayed refs if the transaction is starting to close
791 wait_transaction_pre_flush(tree_root->fs_info);
792 trans = btrfs_start_transaction(tree_root, 1);
795 * we've joined a transaction, make sure it isn't
798 if (trans->transaction->delayed_refs.flushing) {
799 btrfs_end_transaction(trans, tree_root);
803 ret = btrfs_drop_snapshot(trans, root);
807 ret = btrfs_update_root(trans, tree_root,
813 nr = trans->blocks_used;
814 ret = btrfs_end_transaction(trans, tree_root);
817 btrfs_btree_balance_dirty(tree_root, nr);
822 ret = btrfs_del_root(trans, tree_root, &root->root_key);
825 nr = trans->blocks_used;
826 ret = btrfs_end_transaction(trans, tree_root);
829 free_extent_buffer(root->node);
830 free_extent_buffer(root->commit_root);
833 btrfs_btree_balance_dirty(tree_root, nr);
839 * new snapshots need to be created at a very specific time in the
840 * transaction commit. This does the actual creation
842 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
843 struct btrfs_fs_info *fs_info,
844 struct btrfs_pending_snapshot *pending)
846 struct btrfs_key key;
847 struct btrfs_root_item *new_root_item;
848 struct btrfs_root *tree_root = fs_info->tree_root;
849 struct btrfs_root *root = pending->root;
850 struct btrfs_root *parent_root;
851 struct inode *parent_inode;
852 struct dentry *dentry;
853 struct extent_buffer *tmp;
854 struct extent_buffer *old;
860 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
861 if (!new_root_item) {
862 pending->error = -ENOMEM;
866 ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
868 pending->error = ret;
872 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
873 btrfs_orphan_pre_snapshot(trans, pending, &to_reserve);
875 if (to_reserve > 0) {
876 ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv,
879 pending->error = ret;
884 key.objectid = objectid;
885 key.offset = (u64)-1;
886 key.type = BTRFS_ROOT_ITEM_KEY;
888 trans->block_rsv = &pending->block_rsv;
890 dentry = pending->dentry;
891 parent_inode = dentry->d_parent->d_inode;
892 parent_root = BTRFS_I(parent_inode)->root;
893 record_root_in_trans(trans, parent_root);
896 * insert the directory item
898 ret = btrfs_set_inode_index(parent_inode, &index);
900 ret = btrfs_insert_dir_item(trans, parent_root,
901 dentry->d_name.name, dentry->d_name.len,
902 parent_inode->i_ino, &key,
903 BTRFS_FT_DIR, index);
906 btrfs_i_size_write(parent_inode, parent_inode->i_size +
907 dentry->d_name.len * 2);
908 ret = btrfs_update_inode(trans, parent_root, parent_inode);
911 record_root_in_trans(trans, root);
912 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
913 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
915 old = btrfs_lock_root_node(root);
916 btrfs_cow_block(trans, root, old, NULL, 0, &old);
917 btrfs_set_lock_blocking(old);
919 btrfs_copy_root(trans, root, old, &tmp, objectid);
920 btrfs_tree_unlock(old);
921 free_extent_buffer(old);
923 btrfs_set_root_node(new_root_item, tmp);
924 /* record when the snapshot was created in key.offset */
925 key.offset = trans->transid;
926 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
927 btrfs_tree_unlock(tmp);
928 free_extent_buffer(tmp);
932 * insert root back/forward references
934 ret = btrfs_add_root_ref(trans, tree_root, objectid,
935 parent_root->root_key.objectid,
936 parent_inode->i_ino, index,
937 dentry->d_name.name, dentry->d_name.len);
940 key.offset = (u64)-1;
941 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
942 BUG_ON(IS_ERR(pending->snap));
944 btrfs_reloc_post_snapshot(trans, pending);
945 btrfs_orphan_post_snapshot(trans, pending);
947 kfree(new_root_item);
948 btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
953 * create all the snapshots we've scheduled for creation
955 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
956 struct btrfs_fs_info *fs_info)
958 struct btrfs_pending_snapshot *pending;
959 struct list_head *head = &trans->transaction->pending_snapshots;
962 list_for_each_entry(pending, head, list) {
963 ret = create_pending_snapshot(trans, fs_info, pending);
969 static void update_super_roots(struct btrfs_root *root)
971 struct btrfs_root_item *root_item;
972 struct btrfs_super_block *super;
974 super = &root->fs_info->super_copy;
976 root_item = &root->fs_info->chunk_root->root_item;
977 super->chunk_root = root_item->bytenr;
978 super->chunk_root_generation = root_item->generation;
979 super->chunk_root_level = root_item->level;
981 root_item = &root->fs_info->tree_root->root_item;
982 super->root = root_item->bytenr;
983 super->generation = root_item->generation;
984 super->root_level = root_item->level;
985 if (super->cache_generation != 0 || btrfs_test_opt(root, SPACE_CACHE))
986 super->cache_generation = root_item->generation;
989 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
992 spin_lock(&info->new_trans_lock);
993 if (info->running_transaction)
994 ret = info->running_transaction->in_commit;
995 spin_unlock(&info->new_trans_lock);
999 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1002 spin_lock(&info->new_trans_lock);
1003 if (info->running_transaction)
1004 ret = info->running_transaction->blocked;
1005 spin_unlock(&info->new_trans_lock);
1009 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1010 struct btrfs_root *root)
1012 unsigned long joined = 0;
1013 unsigned long timeout = 1;
1014 struct btrfs_transaction *cur_trans;
1015 struct btrfs_transaction *prev_trans = NULL;
1018 int should_grow = 0;
1019 unsigned long now = get_seconds();
1020 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
1022 btrfs_run_ordered_operations(root, 0);
1024 /* make a pass through all the delayed refs we have so far
1025 * any runnings procs may add more while we are here
1027 ret = btrfs_run_delayed_refs(trans, root, 0);
1030 btrfs_trans_release_metadata(trans, root);
1032 cur_trans = trans->transaction;
1034 * set the flushing flag so procs in this transaction have to
1035 * start sending their work down.
1037 cur_trans->delayed_refs.flushing = 1;
1039 ret = btrfs_run_delayed_refs(trans, root, 0);
1042 mutex_lock(&root->fs_info->trans_mutex);
1043 if (cur_trans->in_commit) {
1044 cur_trans->use_count++;
1045 mutex_unlock(&root->fs_info->trans_mutex);
1046 btrfs_end_transaction(trans, root);
1048 ret = wait_for_commit(root, cur_trans);
1051 mutex_lock(&root->fs_info->trans_mutex);
1052 put_transaction(cur_trans);
1053 mutex_unlock(&root->fs_info->trans_mutex);
1058 trans->transaction->in_commit = 1;
1059 trans->transaction->blocked = 1;
1060 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1061 prev_trans = list_entry(cur_trans->list.prev,
1062 struct btrfs_transaction, list);
1063 if (!prev_trans->commit_done) {
1064 prev_trans->use_count++;
1065 mutex_unlock(&root->fs_info->trans_mutex);
1067 wait_for_commit(root, prev_trans);
1069 mutex_lock(&root->fs_info->trans_mutex);
1070 put_transaction(prev_trans);
1074 if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
1078 int snap_pending = 0;
1079 joined = cur_trans->num_joined;
1080 if (!list_empty(&trans->transaction->pending_snapshots))
1083 WARN_ON(cur_trans != trans->transaction);
1084 if (cur_trans->num_writers > 1)
1085 timeout = MAX_SCHEDULE_TIMEOUT;
1086 else if (should_grow)
1089 mutex_unlock(&root->fs_info->trans_mutex);
1091 if (flush_on_commit || snap_pending) {
1092 btrfs_start_delalloc_inodes(root, 1);
1093 ret = btrfs_wait_ordered_extents(root, 0, 1);
1098 * rename don't use btrfs_join_transaction, so, once we
1099 * set the transaction to blocked above, we aren't going
1100 * to get any new ordered operations. We can safely run
1101 * it here and no for sure that nothing new will be added
1104 btrfs_run_ordered_operations(root, 1);
1106 prepare_to_wait(&cur_trans->writer_wait, &wait,
1107 TASK_UNINTERRUPTIBLE);
1110 if (cur_trans->num_writers > 1 || should_grow)
1111 schedule_timeout(timeout);
1113 mutex_lock(&root->fs_info->trans_mutex);
1114 finish_wait(&cur_trans->writer_wait, &wait);
1115 } while (cur_trans->num_writers > 1 ||
1116 (should_grow && cur_trans->num_joined != joined));
1118 ret = create_pending_snapshots(trans, root->fs_info);
1121 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1124 WARN_ON(cur_trans != trans->transaction);
1126 /* btrfs_commit_tree_roots is responsible for getting the
1127 * various roots consistent with each other. Every pointer
1128 * in the tree of tree roots has to point to the most up to date
1129 * root for every subvolume and other tree. So, we have to keep
1130 * the tree logging code from jumping in and changing any
1133 * At this point in the commit, there can't be any tree-log
1134 * writers, but a little lower down we drop the trans mutex
1135 * and let new people in. By holding the tree_log_mutex
1136 * from now until after the super is written, we avoid races
1137 * with the tree-log code.
1139 mutex_lock(&root->fs_info->tree_log_mutex);
1141 ret = commit_fs_roots(trans, root);
1144 /* commit_fs_roots gets rid of all the tree log roots, it is now
1145 * safe to free the root of tree log roots
1147 btrfs_free_log_root_tree(trans, root->fs_info);
1149 ret = commit_cowonly_roots(trans, root);
1152 btrfs_prepare_extent_commit(trans, root);
1154 cur_trans = root->fs_info->running_transaction;
1155 spin_lock(&root->fs_info->new_trans_lock);
1156 root->fs_info->running_transaction = NULL;
1157 spin_unlock(&root->fs_info->new_trans_lock);
1159 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1160 root->fs_info->tree_root->node);
1161 switch_commit_root(root->fs_info->tree_root);
1163 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1164 root->fs_info->chunk_root->node);
1165 switch_commit_root(root->fs_info->chunk_root);
1167 update_super_roots(root);
1169 if (!root->fs_info->log_root_recovering) {
1170 btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
1171 btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
1174 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
1175 sizeof(root->fs_info->super_copy));
1177 trans->transaction->blocked = 0;
1179 wake_up(&root->fs_info->transaction_wait);
1181 mutex_unlock(&root->fs_info->trans_mutex);
1182 ret = btrfs_write_and_wait_transaction(trans, root);
1184 write_ctree_super(trans, root, 0);
1187 * the super is written, we can safely allow the tree-loggers
1188 * to go about their business
1190 mutex_unlock(&root->fs_info->tree_log_mutex);
1192 btrfs_finish_extent_commit(trans, root);
1194 mutex_lock(&root->fs_info->trans_mutex);
1196 cur_trans->commit_done = 1;
1198 root->fs_info->last_trans_committed = cur_trans->transid;
1200 wake_up(&cur_trans->commit_wait);
1202 put_transaction(cur_trans);
1203 put_transaction(cur_trans);
1205 mutex_unlock(&root->fs_info->trans_mutex);
1207 if (current->journal_info == trans)
1208 current->journal_info = NULL;
1210 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1212 if (current != root->fs_info->transaction_kthread)
1213 btrfs_run_delayed_iputs(root);
1219 * interface function to delete all the snapshots we have scheduled for deletion
1221 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1224 struct btrfs_fs_info *fs_info = root->fs_info;
1226 mutex_lock(&fs_info->trans_mutex);
1227 list_splice_init(&fs_info->dead_roots, &list);
1228 mutex_unlock(&fs_info->trans_mutex);
1230 while (!list_empty(&list)) {
1231 root = list_entry(list.next, struct btrfs_root, root_list);
1232 list_del(&root->root_list);
1234 if (btrfs_header_backref_rev(root->node) <
1235 BTRFS_MIXED_BACKREF_REV)
1236 btrfs_drop_snapshot(root, NULL, 0);
1238 btrfs_drop_snapshot(root, NULL, 1);