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 struct btrfs_trans_handle
*start_transaction(struct btrfs_root
*root
,
169 int num_blocks
, int type
)
171 struct btrfs_trans_handle
*h
=
172 kmem_cache_alloc(btrfs_trans_handle_cachep
, GFP_NOFS
);
175 mutex_lock(&root
->fs_info
->trans_mutex
);
176 if (!root
->fs_info
->log_root_recovering
&&
177 ((type
== TRANS_START
&& !root
->fs_info
->open_ioctl_trans
) ||
178 type
== TRANS_USERSPACE
))
179 wait_current_trans(root
);
180 ret
= join_transaction(root
);
183 h
->transid
= root
->fs_info
->running_transaction
->transid
;
184 h
->transaction
= root
->fs_info
->running_transaction
;
185 h
->blocks_reserved
= num_blocks
;
188 h
->delayed_ref_updates
= 0;
191 if (!current
->journal_info
&& type
!= TRANS_USERSPACE
)
192 current
->journal_info
= h
;
194 root
->fs_info
->running_transaction
->use_count
++;
195 record_root_in_trans(h
, root
);
196 mutex_unlock(&root
->fs_info
->trans_mutex
);
200 struct btrfs_trans_handle
*btrfs_start_transaction(struct btrfs_root
*root
,
203 return start_transaction(root
, num_blocks
, TRANS_START
);
205 struct btrfs_trans_handle
*btrfs_join_transaction(struct btrfs_root
*root
,
208 return start_transaction(root
, num_blocks
, TRANS_JOIN
);
211 struct btrfs_trans_handle
*btrfs_start_ioctl_transaction(struct btrfs_root
*r
,
214 return start_transaction(r
, num_blocks
, TRANS_USERSPACE
);
217 /* wait for a transaction commit to be fully complete */
218 static noinline
int wait_for_commit(struct btrfs_root
*root
,
219 struct btrfs_transaction
*commit
)
222 mutex_lock(&root
->fs_info
->trans_mutex
);
223 while (!commit
->commit_done
) {
224 prepare_to_wait(&commit
->commit_wait
, &wait
,
225 TASK_UNINTERRUPTIBLE
);
226 if (commit
->commit_done
)
228 mutex_unlock(&root
->fs_info
->trans_mutex
);
230 mutex_lock(&root
->fs_info
->trans_mutex
);
232 mutex_unlock(&root
->fs_info
->trans_mutex
);
233 finish_wait(&commit
->commit_wait
, &wait
);
239 * rate limit against the drop_snapshot code. This helps to slow down new
240 * operations if the drop_snapshot code isn't able to keep up.
242 static void throttle_on_drops(struct btrfs_root
*root
)
244 struct btrfs_fs_info
*info
= root
->fs_info
;
245 int harder_count
= 0;
248 if (atomic_read(&info
->throttles
)) {
251 thr
= atomic_read(&info
->throttle_gen
);
254 prepare_to_wait(&info
->transaction_throttle
,
255 &wait
, TASK_UNINTERRUPTIBLE
);
256 if (!atomic_read(&info
->throttles
)) {
257 finish_wait(&info
->transaction_throttle
, &wait
);
261 finish_wait(&info
->transaction_throttle
, &wait
);
262 } while (thr
== atomic_read(&info
->throttle_gen
));
265 if (root
->fs_info
->total_ref_cache_size
> 1 * 1024 * 1024 &&
269 if (root
->fs_info
->total_ref_cache_size
> 5 * 1024 * 1024 &&
273 if (root
->fs_info
->total_ref_cache_size
> 10 * 1024 * 1024 &&
280 void btrfs_throttle(struct btrfs_root
*root
)
282 mutex_lock(&root
->fs_info
->trans_mutex
);
283 if (!root
->fs_info
->open_ioctl_trans
)
284 wait_current_trans(root
);
285 mutex_unlock(&root
->fs_info
->trans_mutex
);
288 static int __btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
289 struct btrfs_root
*root
, int throttle
)
291 struct btrfs_transaction
*cur_trans
;
292 struct btrfs_fs_info
*info
= root
->fs_info
;
296 unsigned long cur
= trans
->delayed_ref_updates
;
297 trans
->delayed_ref_updates
= 0;
299 trans
->transaction
->delayed_refs
.num_heads_ready
> 64) {
300 trans
->delayed_ref_updates
= 0;
303 * do a full flush if the transaction is trying
306 if (trans
->transaction
->delayed_refs
.flushing
)
308 btrfs_run_delayed_refs(trans
, root
, cur
);
315 mutex_lock(&info
->trans_mutex
);
316 cur_trans
= info
->running_transaction
;
317 WARN_ON(cur_trans
!= trans
->transaction
);
318 WARN_ON(cur_trans
->num_writers
< 1);
319 cur_trans
->num_writers
--;
321 if (waitqueue_active(&cur_trans
->writer_wait
))
322 wake_up(&cur_trans
->writer_wait
);
323 put_transaction(cur_trans
);
324 mutex_unlock(&info
->trans_mutex
);
326 if (current
->journal_info
== trans
)
327 current
->journal_info
= NULL
;
328 memset(trans
, 0, sizeof(*trans
));
329 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
332 btrfs_run_delayed_iputs(root
);
337 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
338 struct btrfs_root
*root
)
340 return __btrfs_end_transaction(trans
, root
, 0);
343 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
,
344 struct btrfs_root
*root
)
346 return __btrfs_end_transaction(trans
, root
, 1);
350 * when btree blocks are allocated, they have some corresponding bits set for
351 * them in one of two extent_io trees. This is used to make sure all of
352 * those extents are sent to disk but does not wait on them
354 int btrfs_write_marked_extents(struct btrfs_root
*root
,
355 struct extent_io_tree
*dirty_pages
, int mark
)
361 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
367 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
371 while (start
<= end
) {
374 index
= start
>> PAGE_CACHE_SHIFT
;
375 start
= (u64
)(index
+ 1) << PAGE_CACHE_SHIFT
;
376 page
= find_get_page(btree_inode
->i_mapping
, index
);
380 btree_lock_page_hook(page
);
381 if (!page
->mapping
) {
383 page_cache_release(page
);
387 if (PageWriteback(page
)) {
389 wait_on_page_writeback(page
);
392 page_cache_release(page
);
396 err
= write_one_page(page
, 0);
399 page_cache_release(page
);
408 * when btree blocks are allocated, they have some corresponding bits set for
409 * them in one of two extent_io trees. This is used to make sure all of
410 * those extents are on disk for transaction or log commit. We wait
411 * on all the pages and clear them from the dirty pages state tree
413 int btrfs_wait_marked_extents(struct btrfs_root
*root
,
414 struct extent_io_tree
*dirty_pages
, int mark
)
420 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
426 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
431 clear_extent_bits(dirty_pages
, start
, end
, mark
, GFP_NOFS
);
432 while (start
<= end
) {
433 index
= start
>> PAGE_CACHE_SHIFT
;
434 start
= (u64
)(index
+ 1) << PAGE_CACHE_SHIFT
;
435 page
= find_get_page(btree_inode
->i_mapping
, index
);
438 if (PageDirty(page
)) {
439 btree_lock_page_hook(page
);
440 wait_on_page_writeback(page
);
441 err
= write_one_page(page
, 0);
445 wait_on_page_writeback(page
);
446 page_cache_release(page
);
456 * when btree blocks are allocated, they have some corresponding bits set for
457 * them in one of two extent_io trees. This is used to make sure all of
458 * those extents are on disk for transaction or log commit
460 int btrfs_write_and_wait_marked_extents(struct btrfs_root
*root
,
461 struct extent_io_tree
*dirty_pages
, int mark
)
466 ret
= btrfs_write_marked_extents(root
, dirty_pages
, mark
);
467 ret2
= btrfs_wait_marked_extents(root
, dirty_pages
, mark
);
471 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
,
472 struct btrfs_root
*root
)
474 if (!trans
|| !trans
->transaction
) {
475 struct inode
*btree_inode
;
476 btree_inode
= root
->fs_info
->btree_inode
;
477 return filemap_write_and_wait(btree_inode
->i_mapping
);
479 return btrfs_write_and_wait_marked_extents(root
,
480 &trans
->transaction
->dirty_pages
,
485 * this is used to update the root pointer in the tree of tree roots.
487 * But, in the case of the extent allocation tree, updating the root
488 * pointer may allocate blocks which may change the root of the extent
491 * So, this loops and repeats and makes sure the cowonly root didn't
492 * change while the root pointer was being updated in the metadata.
494 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
495 struct btrfs_root
*root
)
500 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
502 old_root_used
= btrfs_root_used(&root
->root_item
);
503 btrfs_write_dirty_block_groups(trans
, root
);
506 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
507 if (old_root_bytenr
== root
->node
->start
&&
508 old_root_used
== btrfs_root_used(&root
->root_item
))
511 btrfs_set_root_node(&root
->root_item
, root
->node
);
512 ret
= btrfs_update_root(trans
, tree_root
,
517 old_root_used
= btrfs_root_used(&root
->root_item
);
518 ret
= btrfs_write_dirty_block_groups(trans
, root
);
522 if (root
!= root
->fs_info
->extent_root
)
523 switch_commit_root(root
);
529 * update all the cowonly tree roots on disk
531 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
,
532 struct btrfs_root
*root
)
534 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
535 struct list_head
*next
;
536 struct extent_buffer
*eb
;
539 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
542 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
543 btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
, 0, &eb
);
544 btrfs_tree_unlock(eb
);
545 free_extent_buffer(eb
);
547 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
550 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
551 next
= fs_info
->dirty_cowonly_roots
.next
;
553 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
555 update_cowonly_root(trans
, root
);
558 down_write(&fs_info
->extent_commit_sem
);
559 switch_commit_root(fs_info
->extent_root
);
560 up_write(&fs_info
->extent_commit_sem
);
566 * dead roots are old snapshots that need to be deleted. This allocates
567 * a dirty root struct and adds it into the list of dead roots that need to
570 int btrfs_add_dead_root(struct btrfs_root
*root
)
572 mutex_lock(&root
->fs_info
->trans_mutex
);
573 list_add(&root
->root_list
, &root
->fs_info
->dead_roots
);
574 mutex_unlock(&root
->fs_info
->trans_mutex
);
579 * update all the cowonly tree roots on disk
581 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
,
582 struct btrfs_root
*root
)
584 struct btrfs_root
*gang
[8];
585 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
591 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
594 BTRFS_ROOT_TRANS_TAG
);
597 for (i
= 0; i
< ret
; i
++) {
599 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
600 (unsigned long)root
->root_key
.objectid
,
601 BTRFS_ROOT_TRANS_TAG
);
603 btrfs_free_log(trans
, root
);
604 btrfs_update_reloc_root(trans
, root
);
606 if (root
->commit_root
!= root
->node
) {
607 switch_commit_root(root
);
608 btrfs_set_root_node(&root
->root_item
,
612 err
= btrfs_update_root(trans
, fs_info
->tree_root
,
623 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
624 * otherwise every leaf in the btree is read and defragged.
626 int btrfs_defrag_root(struct btrfs_root
*root
, int cacheonly
)
628 struct btrfs_fs_info
*info
= root
->fs_info
;
630 struct btrfs_trans_handle
*trans
;
634 if (root
->defrag_running
)
636 trans
= btrfs_start_transaction(root
, 1);
638 root
->defrag_running
= 1;
639 ret
= btrfs_defrag_leaves(trans
, root
, cacheonly
);
640 nr
= trans
->blocks_used
;
641 btrfs_end_transaction(trans
, root
);
642 btrfs_btree_balance_dirty(info
->tree_root
, nr
);
645 trans
= btrfs_start_transaction(root
, 1);
646 if (root
->fs_info
->closing
|| ret
!= -EAGAIN
)
649 root
->defrag_running
= 0;
651 btrfs_end_transaction(trans
, root
);
657 * when dropping snapshots, we generate a ton of delayed refs, and it makes
658 * sense not to join the transaction while it is trying to flush the current
659 * queue of delayed refs out.
661 * This is used by the drop snapshot code only
663 static noinline
int wait_transaction_pre_flush(struct btrfs_fs_info
*info
)
667 mutex_lock(&info
->trans_mutex
);
668 while (info
->running_transaction
&&
669 info
->running_transaction
->delayed_refs
.flushing
) {
670 prepare_to_wait(&info
->transaction_wait
, &wait
,
671 TASK_UNINTERRUPTIBLE
);
672 mutex_unlock(&info
->trans_mutex
);
676 mutex_lock(&info
->trans_mutex
);
677 finish_wait(&info
->transaction_wait
, &wait
);
679 mutex_unlock(&info
->trans_mutex
);
684 * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
687 int btrfs_drop_dead_root(struct btrfs_root
*root
)
689 struct btrfs_trans_handle
*trans
;
690 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
696 * we don't want to jump in and create a bunch of
697 * delayed refs if the transaction is starting to close
699 wait_transaction_pre_flush(tree_root
->fs_info
);
700 trans
= btrfs_start_transaction(tree_root
, 1);
703 * we've joined a transaction, make sure it isn't
706 if (trans
->transaction
->delayed_refs
.flushing
) {
707 btrfs_end_transaction(trans
, tree_root
);
711 ret
= btrfs_drop_snapshot(trans
, root
);
715 ret
= btrfs_update_root(trans
, tree_root
,
721 nr
= trans
->blocks_used
;
722 ret
= btrfs_end_transaction(trans
, tree_root
);
725 btrfs_btree_balance_dirty(tree_root
, nr
);
730 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
733 nr
= trans
->blocks_used
;
734 ret
= btrfs_end_transaction(trans
, tree_root
);
737 free_extent_buffer(root
->node
);
738 free_extent_buffer(root
->commit_root
);
741 btrfs_btree_balance_dirty(tree_root
, nr
);
747 * new snapshots need to be created at a very specific time in the
748 * transaction commit. This does the actual creation
750 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
751 struct btrfs_fs_info
*fs_info
,
752 struct btrfs_pending_snapshot
*pending
)
754 struct btrfs_key key
;
755 struct btrfs_root_item
*new_root_item
;
756 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
757 struct btrfs_root
*root
= pending
->root
;
758 struct btrfs_root
*parent_root
;
759 struct inode
*parent_inode
;
760 struct extent_buffer
*tmp
;
761 struct extent_buffer
*old
;
767 parent_inode
= pending
->dentry
->d_parent
->d_inode
;
768 parent_root
= BTRFS_I(parent_inode
)->root
;
770 new_root_item
= kmalloc(sizeof(*new_root_item
), GFP_NOFS
);
771 if (!new_root_item
) {
775 ret
= btrfs_find_free_objectid(trans
, tree_root
, 0, &objectid
);
779 key
.objectid
= objectid
;
780 /* record when the snapshot was created in key.offset */
781 key
.offset
= trans
->transid
;
782 btrfs_set_key_type(&key
, BTRFS_ROOT_ITEM_KEY
);
784 memcpy(&pending
->root_key
, &key
, sizeof(key
));
785 pending
->root_key
.offset
= (u64
)-1;
787 record_root_in_trans(trans
, parent_root
);
789 * insert the directory item
791 namelen
= strlen(pending
->name
);
792 ret
= btrfs_set_inode_index(parent_inode
, &index
);
794 ret
= btrfs_insert_dir_item(trans
, parent_root
,
795 pending
->name
, namelen
,
797 &pending
->root_key
, BTRFS_FT_DIR
, index
);
800 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+ namelen
* 2);
801 ret
= btrfs_update_inode(trans
, parent_root
, parent_inode
);
804 record_root_in_trans(trans
, root
);
805 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
806 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
808 old
= btrfs_lock_root_node(root
);
809 btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
);
810 btrfs_set_lock_blocking(old
);
812 btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
813 btrfs_tree_unlock(old
);
814 free_extent_buffer(old
);
816 btrfs_set_root_node(new_root_item
, tmp
);
817 ret
= btrfs_insert_root(trans
, root
->fs_info
->tree_root
, &key
,
820 btrfs_tree_unlock(tmp
);
821 free_extent_buffer(tmp
);
823 ret
= btrfs_add_root_ref(trans
, parent_root
->fs_info
->tree_root
,
824 pending
->root_key
.objectid
,
825 parent_root
->root_key
.objectid
,
826 parent_inode
->i_ino
, index
, pending
->name
,
831 kfree(new_root_item
);
836 * create all the snapshots we've scheduled for creation
838 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
,
839 struct btrfs_fs_info
*fs_info
)
841 struct btrfs_pending_snapshot
*pending
;
842 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
845 list_for_each_entry(pending
, head
, list
) {
846 ret
= create_pending_snapshot(trans
, fs_info
, pending
);
852 static void update_super_roots(struct btrfs_root
*root
)
854 struct btrfs_root_item
*root_item
;
855 struct btrfs_super_block
*super
;
857 super
= &root
->fs_info
->super_copy
;
859 root_item
= &root
->fs_info
->chunk_root
->root_item
;
860 super
->chunk_root
= root_item
->bytenr
;
861 super
->chunk_root_generation
= root_item
->generation
;
862 super
->chunk_root_level
= root_item
->level
;
864 root_item
= &root
->fs_info
->tree_root
->root_item
;
865 super
->root
= root_item
->bytenr
;
866 super
->generation
= root_item
->generation
;
867 super
->root_level
= root_item
->level
;
870 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
873 spin_lock(&info
->new_trans_lock
);
874 if (info
->running_transaction
)
875 ret
= info
->running_transaction
->in_commit
;
876 spin_unlock(&info
->new_trans_lock
);
880 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
,
881 struct btrfs_root
*root
)
883 unsigned long joined
= 0;
884 unsigned long timeout
= 1;
885 struct btrfs_transaction
*cur_trans
;
886 struct btrfs_transaction
*prev_trans
= NULL
;
890 unsigned long now
= get_seconds();
891 int flush_on_commit
= btrfs_test_opt(root
, FLUSHONCOMMIT
);
893 btrfs_run_ordered_operations(root
, 0);
895 /* make a pass through all the delayed refs we have so far
896 * any runnings procs may add more while we are here
898 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
901 cur_trans
= trans
->transaction
;
903 * set the flushing flag so procs in this transaction have to
904 * start sending their work down.
906 cur_trans
->delayed_refs
.flushing
= 1;
908 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
911 mutex_lock(&root
->fs_info
->trans_mutex
);
912 if (cur_trans
->in_commit
) {
913 cur_trans
->use_count
++;
914 mutex_unlock(&root
->fs_info
->trans_mutex
);
915 btrfs_end_transaction(trans
, root
);
917 ret
= wait_for_commit(root
, cur_trans
);
920 mutex_lock(&root
->fs_info
->trans_mutex
);
921 put_transaction(cur_trans
);
922 mutex_unlock(&root
->fs_info
->trans_mutex
);
927 trans
->transaction
->in_commit
= 1;
928 trans
->transaction
->blocked
= 1;
929 if (cur_trans
->list
.prev
!= &root
->fs_info
->trans_list
) {
930 prev_trans
= list_entry(cur_trans
->list
.prev
,
931 struct btrfs_transaction
, list
);
932 if (!prev_trans
->commit_done
) {
933 prev_trans
->use_count
++;
934 mutex_unlock(&root
->fs_info
->trans_mutex
);
936 wait_for_commit(root
, prev_trans
);
938 mutex_lock(&root
->fs_info
->trans_mutex
);
939 put_transaction(prev_trans
);
943 if (now
< cur_trans
->start_time
|| now
- cur_trans
->start_time
< 1)
947 int snap_pending
= 0;
948 joined
= cur_trans
->num_joined
;
949 if (!list_empty(&trans
->transaction
->pending_snapshots
))
952 WARN_ON(cur_trans
!= trans
->transaction
);
953 prepare_to_wait(&cur_trans
->writer_wait
, &wait
,
954 TASK_UNINTERRUPTIBLE
);
956 if (cur_trans
->num_writers
> 1)
957 timeout
= MAX_SCHEDULE_TIMEOUT
;
958 else if (should_grow
)
961 mutex_unlock(&root
->fs_info
->trans_mutex
);
963 if (flush_on_commit
|| snap_pending
) {
964 btrfs_start_delalloc_inodes(root
, 1);
965 ret
= btrfs_wait_ordered_extents(root
, 0, 1);
970 * rename don't use btrfs_join_transaction, so, once we
971 * set the transaction to blocked above, we aren't going
972 * to get any new ordered operations. We can safely run
973 * it here and no for sure that nothing new will be added
976 btrfs_run_ordered_operations(root
, 1);
979 if (cur_trans
->num_writers
> 1 || should_grow
)
980 schedule_timeout(timeout
);
982 mutex_lock(&root
->fs_info
->trans_mutex
);
983 finish_wait(&cur_trans
->writer_wait
, &wait
);
984 } while (cur_trans
->num_writers
> 1 ||
985 (should_grow
&& cur_trans
->num_joined
!= joined
));
987 ret
= create_pending_snapshots(trans
, root
->fs_info
);
990 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
993 WARN_ON(cur_trans
!= trans
->transaction
);
995 /* btrfs_commit_tree_roots is responsible for getting the
996 * various roots consistent with each other. Every pointer
997 * in the tree of tree roots has to point to the most up to date
998 * root for every subvolume and other tree. So, we have to keep
999 * the tree logging code from jumping in and changing any
1002 * At this point in the commit, there can't be any tree-log
1003 * writers, but a little lower down we drop the trans mutex
1004 * and let new people in. By holding the tree_log_mutex
1005 * from now until after the super is written, we avoid races
1006 * with the tree-log code.
1008 mutex_lock(&root
->fs_info
->tree_log_mutex
);
1010 ret
= commit_fs_roots(trans
, root
);
1013 /* commit_fs_roots gets rid of all the tree log roots, it is now
1014 * safe to free the root of tree log roots
1016 btrfs_free_log_root_tree(trans
, root
->fs_info
);
1018 ret
= commit_cowonly_roots(trans
, root
);
1021 btrfs_prepare_extent_commit(trans
, root
);
1023 cur_trans
= root
->fs_info
->running_transaction
;
1024 spin_lock(&root
->fs_info
->new_trans_lock
);
1025 root
->fs_info
->running_transaction
= NULL
;
1026 spin_unlock(&root
->fs_info
->new_trans_lock
);
1028 btrfs_set_root_node(&root
->fs_info
->tree_root
->root_item
,
1029 root
->fs_info
->tree_root
->node
);
1030 switch_commit_root(root
->fs_info
->tree_root
);
1032 btrfs_set_root_node(&root
->fs_info
->chunk_root
->root_item
,
1033 root
->fs_info
->chunk_root
->node
);
1034 switch_commit_root(root
->fs_info
->chunk_root
);
1036 update_super_roots(root
);
1038 if (!root
->fs_info
->log_root_recovering
) {
1039 btrfs_set_super_log_root(&root
->fs_info
->super_copy
, 0);
1040 btrfs_set_super_log_root_level(&root
->fs_info
->super_copy
, 0);
1043 memcpy(&root
->fs_info
->super_for_commit
, &root
->fs_info
->super_copy
,
1044 sizeof(root
->fs_info
->super_copy
));
1046 trans
->transaction
->blocked
= 0;
1048 wake_up(&root
->fs_info
->transaction_wait
);
1050 mutex_unlock(&root
->fs_info
->trans_mutex
);
1051 ret
= btrfs_write_and_wait_transaction(trans
, root
);
1053 write_ctree_super(trans
, root
, 0);
1056 * the super is written, we can safely allow the tree-loggers
1057 * to go about their business
1059 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1061 btrfs_finish_extent_commit(trans
, root
);
1063 mutex_lock(&root
->fs_info
->trans_mutex
);
1065 cur_trans
->commit_done
= 1;
1067 root
->fs_info
->last_trans_committed
= cur_trans
->transid
;
1069 wake_up(&cur_trans
->commit_wait
);
1071 put_transaction(cur_trans
);
1072 put_transaction(cur_trans
);
1074 mutex_unlock(&root
->fs_info
->trans_mutex
);
1076 if (current
->journal_info
== trans
)
1077 current
->journal_info
= NULL
;
1079 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1081 if (current
!= root
->fs_info
->transaction_kthread
)
1082 btrfs_run_delayed_iputs(root
);
1088 * interface function to delete all the snapshots we have scheduled for deletion
1090 int btrfs_clean_old_snapshots(struct btrfs_root
*root
)
1093 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1095 mutex_lock(&fs_info
->trans_mutex
);
1096 list_splice_init(&fs_info
->dead_roots
, &list
);
1097 mutex_unlock(&fs_info
->trans_mutex
);
1099 while (!list_empty(&list
)) {
1100 root
= list_entry(list
.next
, struct btrfs_root
, root_list
);
1101 list_del(&root
->root_list
);
1103 if (btrfs_header_backref_rev(root
->node
) <
1104 BTRFS_MIXED_BACKREF_REV
)
1105 btrfs_drop_snapshot(root
, 0);
1107 btrfs_drop_snapshot(root
, 1);