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"
30 #include "inode-map.h"
32 #define BTRFS_ROOT_TRANS_TAG 0
34 static noinline
void put_transaction(struct btrfs_transaction
*transaction
)
36 WARN_ON(atomic_read(&transaction
->use_count
) == 0);
37 if (atomic_dec_and_test(&transaction
->use_count
)) {
38 memset(transaction
, 0, sizeof(*transaction
));
39 kmem_cache_free(btrfs_transaction_cachep
, transaction
);
43 static noinline
void switch_commit_root(struct btrfs_root
*root
)
45 free_extent_buffer(root
->commit_root
);
46 root
->commit_root
= btrfs_root_node(root
);
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 atomic_set(&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 atomic_set(&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 atomic_inc(&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 atomic_inc(&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
;
186 if (root
->fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
)
187 return ERR_PTR(-EROFS
);
189 h
= kmem_cache_alloc(btrfs_trans_handle_cachep
, GFP_NOFS
);
191 return ERR_PTR(-ENOMEM
);
193 if (type
!= TRANS_JOIN_NOLOCK
)
194 mutex_lock(&root
->fs_info
->trans_mutex
);
195 if (may_wait_transaction(root
, type
))
196 wait_current_trans(root
);
198 ret
= join_transaction(root
);
200 kmem_cache_free(btrfs_trans_handle_cachep
, h
);
201 if (type
!= TRANS_JOIN_NOLOCK
)
202 mutex_unlock(&root
->fs_info
->trans_mutex
);
206 cur_trans
= root
->fs_info
->running_transaction
;
207 atomic_inc(&cur_trans
->use_count
);
208 if (type
!= TRANS_JOIN_NOLOCK
)
209 mutex_unlock(&root
->fs_info
->trans_mutex
);
211 h
->transid
= cur_trans
->transid
;
212 h
->transaction
= cur_trans
;
215 h
->bytes_reserved
= 0;
216 h
->delayed_ref_updates
= 0;
220 if (cur_trans
->blocked
&& may_wait_transaction(root
, type
)) {
221 btrfs_commit_transaction(h
, root
);
226 ret
= btrfs_trans_reserve_metadata(h
, root
, num_items
);
227 if (ret
== -EAGAIN
&& !retries
) {
229 btrfs_commit_transaction(h
, root
);
231 } else if (ret
== -EAGAIN
) {
233 * We have already retried and got EAGAIN, so really we
234 * don't have space, so set ret to -ENOSPC.
240 btrfs_end_transaction(h
, root
);
245 if (type
!= TRANS_JOIN_NOLOCK
)
246 mutex_lock(&root
->fs_info
->trans_mutex
);
247 record_root_in_trans(h
, root
);
248 if (type
!= TRANS_JOIN_NOLOCK
)
249 mutex_unlock(&root
->fs_info
->trans_mutex
);
251 if (!current
->journal_info
&& type
!= TRANS_USERSPACE
)
252 current
->journal_info
= h
;
256 struct btrfs_trans_handle
*btrfs_start_transaction(struct btrfs_root
*root
,
259 return start_transaction(root
, num_items
, TRANS_START
);
261 struct btrfs_trans_handle
*btrfs_join_transaction(struct btrfs_root
*root
,
264 return start_transaction(root
, 0, TRANS_JOIN
);
267 struct btrfs_trans_handle
*btrfs_join_transaction_nolock(struct btrfs_root
*root
,
270 return start_transaction(root
, 0, TRANS_JOIN_NOLOCK
);
273 struct btrfs_trans_handle
*btrfs_start_ioctl_transaction(struct btrfs_root
*r
,
276 return start_transaction(r
, 0, TRANS_USERSPACE
);
279 /* wait for a transaction commit to be fully complete */
280 static noinline
int wait_for_commit(struct btrfs_root
*root
,
281 struct btrfs_transaction
*commit
)
284 mutex_lock(&root
->fs_info
->trans_mutex
);
285 while (!commit
->commit_done
) {
286 prepare_to_wait(&commit
->commit_wait
, &wait
,
287 TASK_UNINTERRUPTIBLE
);
288 if (commit
->commit_done
)
290 mutex_unlock(&root
->fs_info
->trans_mutex
);
292 mutex_lock(&root
->fs_info
->trans_mutex
);
294 mutex_unlock(&root
->fs_info
->trans_mutex
);
295 finish_wait(&commit
->commit_wait
, &wait
);
299 int btrfs_wait_for_commit(struct btrfs_root
*root
, u64 transid
)
301 struct btrfs_transaction
*cur_trans
= NULL
, *t
;
304 mutex_lock(&root
->fs_info
->trans_mutex
);
308 if (transid
<= root
->fs_info
->last_trans_committed
)
311 /* find specified transaction */
312 list_for_each_entry(t
, &root
->fs_info
->trans_list
, list
) {
313 if (t
->transid
== transid
) {
317 if (t
->transid
> transid
)
322 goto out_unlock
; /* bad transid */
324 /* find newest transaction that is committing | committed */
325 list_for_each_entry_reverse(t
, &root
->fs_info
->trans_list
,
335 goto out_unlock
; /* nothing committing|committed */
338 atomic_inc(&cur_trans
->use_count
);
339 mutex_unlock(&root
->fs_info
->trans_mutex
);
341 wait_for_commit(root
, cur_trans
);
343 mutex_lock(&root
->fs_info
->trans_mutex
);
344 put_transaction(cur_trans
);
347 mutex_unlock(&root
->fs_info
->trans_mutex
);
353 * rate limit against the drop_snapshot code. This helps to slow down new
354 * operations if the drop_snapshot code isn't able to keep up.
356 static void throttle_on_drops(struct btrfs_root
*root
)
358 struct btrfs_fs_info
*info
= root
->fs_info
;
359 int harder_count
= 0;
362 if (atomic_read(&info
->throttles
)) {
365 thr
= atomic_read(&info
->throttle_gen
);
368 prepare_to_wait(&info
->transaction_throttle
,
369 &wait
, TASK_UNINTERRUPTIBLE
);
370 if (!atomic_read(&info
->throttles
)) {
371 finish_wait(&info
->transaction_throttle
, &wait
);
375 finish_wait(&info
->transaction_throttle
, &wait
);
376 } while (thr
== atomic_read(&info
->throttle_gen
));
379 if (root
->fs_info
->total_ref_cache_size
> 1 * 1024 * 1024 &&
383 if (root
->fs_info
->total_ref_cache_size
> 5 * 1024 * 1024 &&
387 if (root
->fs_info
->total_ref_cache_size
> 10 * 1024 * 1024 &&
394 void btrfs_throttle(struct btrfs_root
*root
)
396 mutex_lock(&root
->fs_info
->trans_mutex
);
397 if (!root
->fs_info
->open_ioctl_trans
)
398 wait_current_trans(root
);
399 mutex_unlock(&root
->fs_info
->trans_mutex
);
402 static int should_end_transaction(struct btrfs_trans_handle
*trans
,
403 struct btrfs_root
*root
)
406 ret
= btrfs_block_rsv_check(trans
, root
,
407 &root
->fs_info
->global_block_rsv
, 0, 5);
411 int btrfs_should_end_transaction(struct btrfs_trans_handle
*trans
,
412 struct btrfs_root
*root
)
414 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
417 if (cur_trans
->blocked
|| cur_trans
->delayed_refs
.flushing
)
420 updates
= trans
->delayed_ref_updates
;
421 trans
->delayed_ref_updates
= 0;
423 btrfs_run_delayed_refs(trans
, root
, updates
);
425 return should_end_transaction(trans
, root
);
428 static int __btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
429 struct btrfs_root
*root
, int throttle
, int lock
)
431 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
432 struct btrfs_fs_info
*info
= root
->fs_info
;
436 unsigned long cur
= trans
->delayed_ref_updates
;
437 trans
->delayed_ref_updates
= 0;
439 trans
->transaction
->delayed_refs
.num_heads_ready
> 64) {
440 trans
->delayed_ref_updates
= 0;
443 * do a full flush if the transaction is trying
446 if (trans
->transaction
->delayed_refs
.flushing
)
448 btrfs_run_delayed_refs(trans
, root
, cur
);
455 btrfs_trans_release_metadata(trans
, root
);
457 if (lock
&& !root
->fs_info
->open_ioctl_trans
&&
458 should_end_transaction(trans
, root
))
459 trans
->transaction
->blocked
= 1;
461 if (lock
&& cur_trans
->blocked
&& !cur_trans
->in_commit
) {
463 return btrfs_commit_transaction(trans
, root
);
465 wake_up_process(info
->transaction_kthread
);
468 WARN_ON(cur_trans
!= info
->running_transaction
);
469 WARN_ON(atomic_read(&cur_trans
->num_writers
) < 1);
470 atomic_dec(&cur_trans
->num_writers
);
473 if (waitqueue_active(&cur_trans
->writer_wait
))
474 wake_up(&cur_trans
->writer_wait
);
475 put_transaction(cur_trans
);
477 if (current
->journal_info
== trans
)
478 current
->journal_info
= NULL
;
479 memset(trans
, 0, sizeof(*trans
));
480 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
483 btrfs_run_delayed_iputs(root
);
488 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
489 struct btrfs_root
*root
)
491 return __btrfs_end_transaction(trans
, root
, 0, 1);
494 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
,
495 struct btrfs_root
*root
)
497 return __btrfs_end_transaction(trans
, root
, 1, 1);
500 int btrfs_end_transaction_nolock(struct btrfs_trans_handle
*trans
,
501 struct btrfs_root
*root
)
503 return __btrfs_end_transaction(trans
, root
, 0, 0);
507 * when btree blocks are allocated, they have some corresponding bits set for
508 * them in one of two extent_io trees. This is used to make sure all of
509 * those extents are sent to disk but does not wait on them
511 int btrfs_write_marked_extents(struct btrfs_root
*root
,
512 struct extent_io_tree
*dirty_pages
, int mark
)
518 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
524 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
528 while (start
<= end
) {
531 index
= start
>> PAGE_CACHE_SHIFT
;
532 start
= (u64
)(index
+ 1) << PAGE_CACHE_SHIFT
;
533 page
= find_get_page(btree_inode
->i_mapping
, index
);
537 btree_lock_page_hook(page
);
538 if (!page
->mapping
) {
540 page_cache_release(page
);
544 if (PageWriteback(page
)) {
546 wait_on_page_writeback(page
);
549 page_cache_release(page
);
553 err
= write_one_page(page
, 0);
556 page_cache_release(page
);
565 * when btree blocks are allocated, they have some corresponding bits set for
566 * them in one of two extent_io trees. This is used to make sure all of
567 * those extents are on disk for transaction or log commit. We wait
568 * on all the pages and clear them from the dirty pages state tree
570 int btrfs_wait_marked_extents(struct btrfs_root
*root
,
571 struct extent_io_tree
*dirty_pages
, int mark
)
577 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
583 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
588 clear_extent_bits(dirty_pages
, start
, end
, mark
, GFP_NOFS
);
589 while (start
<= end
) {
590 index
= start
>> PAGE_CACHE_SHIFT
;
591 start
= (u64
)(index
+ 1) << PAGE_CACHE_SHIFT
;
592 page
= find_get_page(btree_inode
->i_mapping
, index
);
595 if (PageDirty(page
)) {
596 btree_lock_page_hook(page
);
597 wait_on_page_writeback(page
);
598 err
= write_one_page(page
, 0);
602 wait_on_page_writeback(page
);
603 page_cache_release(page
);
613 * when btree blocks are allocated, they have some corresponding bits set for
614 * them in one of two extent_io trees. This is used to make sure all of
615 * those extents are on disk for transaction or log commit
617 int btrfs_write_and_wait_marked_extents(struct btrfs_root
*root
,
618 struct extent_io_tree
*dirty_pages
, int mark
)
623 ret
= btrfs_write_marked_extents(root
, dirty_pages
, mark
);
624 ret2
= btrfs_wait_marked_extents(root
, dirty_pages
, mark
);
628 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
,
629 struct btrfs_root
*root
)
631 if (!trans
|| !trans
->transaction
) {
632 struct inode
*btree_inode
;
633 btree_inode
= root
->fs_info
->btree_inode
;
634 return filemap_write_and_wait(btree_inode
->i_mapping
);
636 return btrfs_write_and_wait_marked_extents(root
,
637 &trans
->transaction
->dirty_pages
,
642 * this is used to update the root pointer in the tree of tree roots.
644 * But, in the case of the extent allocation tree, updating the root
645 * pointer may allocate blocks which may change the root of the extent
648 * So, this loops and repeats and makes sure the cowonly root didn't
649 * change while the root pointer was being updated in the metadata.
651 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
652 struct btrfs_root
*root
)
657 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
659 old_root_used
= btrfs_root_used(&root
->root_item
);
660 btrfs_write_dirty_block_groups(trans
, root
);
663 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
664 if (old_root_bytenr
== root
->node
->start
&&
665 old_root_used
== btrfs_root_used(&root
->root_item
))
668 btrfs_set_root_node(&root
->root_item
, root
->node
);
669 ret
= btrfs_update_root(trans
, tree_root
,
674 old_root_used
= btrfs_root_used(&root
->root_item
);
675 ret
= btrfs_write_dirty_block_groups(trans
, root
);
679 if (root
!= root
->fs_info
->extent_root
)
680 switch_commit_root(root
);
686 * update all the cowonly tree roots on disk
688 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
,
689 struct btrfs_root
*root
)
691 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
692 struct list_head
*next
;
693 struct extent_buffer
*eb
;
696 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
699 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
700 btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
, 0, &eb
);
701 btrfs_tree_unlock(eb
);
702 free_extent_buffer(eb
);
704 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
707 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
708 next
= fs_info
->dirty_cowonly_roots
.next
;
710 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
712 update_cowonly_root(trans
, root
);
715 down_write(&fs_info
->extent_commit_sem
);
716 switch_commit_root(fs_info
->extent_root
);
717 up_write(&fs_info
->extent_commit_sem
);
723 * dead roots are old snapshots that need to be deleted. This allocates
724 * a dirty root struct and adds it into the list of dead roots that need to
727 int btrfs_add_dead_root(struct btrfs_root
*root
)
729 mutex_lock(&root
->fs_info
->trans_mutex
);
730 list_add(&root
->root_list
, &root
->fs_info
->dead_roots
);
731 mutex_unlock(&root
->fs_info
->trans_mutex
);
736 * update all the cowonly tree roots on disk
738 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
,
739 struct btrfs_root
*root
)
741 struct btrfs_root
*gang
[8];
742 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
748 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
751 BTRFS_ROOT_TRANS_TAG
);
754 for (i
= 0; i
< ret
; i
++) {
756 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
757 (unsigned long)root
->root_key
.objectid
,
758 BTRFS_ROOT_TRANS_TAG
);
760 btrfs_free_log(trans
, root
);
761 btrfs_update_reloc_root(trans
, root
);
762 btrfs_orphan_commit_root(trans
, root
);
764 if (root
->commit_root
!= root
->node
) {
765 mutex_lock(&root
->fs_commit_mutex
);
766 switch_commit_root(root
);
767 btrfs_unpin_free_ino(root
);
768 mutex_unlock(&root
->fs_commit_mutex
);
770 btrfs_set_root_node(&root
->root_item
,
774 err
= btrfs_update_root(trans
, fs_info
->tree_root
,
785 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
786 * otherwise every leaf in the btree is read and defragged.
788 int btrfs_defrag_root(struct btrfs_root
*root
, int cacheonly
)
790 struct btrfs_fs_info
*info
= root
->fs_info
;
791 struct btrfs_trans_handle
*trans
;
795 if (xchg(&root
->defrag_running
, 1))
799 trans
= btrfs_start_transaction(root
, 0);
801 return PTR_ERR(trans
);
803 ret
= btrfs_defrag_leaves(trans
, root
, cacheonly
);
805 nr
= trans
->blocks_used
;
806 btrfs_end_transaction(trans
, root
);
807 btrfs_btree_balance_dirty(info
->tree_root
, nr
);
810 if (root
->fs_info
->closing
|| ret
!= -EAGAIN
)
813 root
->defrag_running
= 0;
819 * when dropping snapshots, we generate a ton of delayed refs, and it makes
820 * sense not to join the transaction while it is trying to flush the current
821 * queue of delayed refs out.
823 * This is used by the drop snapshot code only
825 static noinline
int wait_transaction_pre_flush(struct btrfs_fs_info
*info
)
829 mutex_lock(&info
->trans_mutex
);
830 while (info
->running_transaction
&&
831 info
->running_transaction
->delayed_refs
.flushing
) {
832 prepare_to_wait(&info
->transaction_wait
, &wait
,
833 TASK_UNINTERRUPTIBLE
);
834 mutex_unlock(&info
->trans_mutex
);
838 mutex_lock(&info
->trans_mutex
);
839 finish_wait(&info
->transaction_wait
, &wait
);
841 mutex_unlock(&info
->trans_mutex
);
846 * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
849 int btrfs_drop_dead_root(struct btrfs_root
*root
)
851 struct btrfs_trans_handle
*trans
;
852 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
858 * we don't want to jump in and create a bunch of
859 * delayed refs if the transaction is starting to close
861 wait_transaction_pre_flush(tree_root
->fs_info
);
862 trans
= btrfs_start_transaction(tree_root
, 1);
865 * we've joined a transaction, make sure it isn't
868 if (trans
->transaction
->delayed_refs
.flushing
) {
869 btrfs_end_transaction(trans
, tree_root
);
873 ret
= btrfs_drop_snapshot(trans
, root
);
877 ret
= btrfs_update_root(trans
, tree_root
,
883 nr
= trans
->blocks_used
;
884 ret
= btrfs_end_transaction(trans
, tree_root
);
887 btrfs_btree_balance_dirty(tree_root
, nr
);
892 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
895 nr
= trans
->blocks_used
;
896 ret
= btrfs_end_transaction(trans
, tree_root
);
899 free_extent_buffer(root
->node
);
900 free_extent_buffer(root
->commit_root
);
903 btrfs_btree_balance_dirty(tree_root
, nr
);
909 * new snapshots need to be created at a very specific time in the
910 * transaction commit. This does the actual creation
912 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
913 struct btrfs_fs_info
*fs_info
,
914 struct btrfs_pending_snapshot
*pending
)
916 struct btrfs_key key
;
917 struct btrfs_root_item
*new_root_item
;
918 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
919 struct btrfs_root
*root
= pending
->root
;
920 struct btrfs_root
*parent_root
;
921 struct inode
*parent_inode
;
922 struct dentry
*parent
;
923 struct dentry
*dentry
;
924 struct extent_buffer
*tmp
;
925 struct extent_buffer
*old
;
932 new_root_item
= kmalloc(sizeof(*new_root_item
), GFP_NOFS
);
933 if (!new_root_item
) {
934 pending
->error
= -ENOMEM
;
938 ret
= btrfs_find_free_objectid(tree_root
, &objectid
);
940 pending
->error
= ret
;
944 btrfs_reloc_pre_snapshot(trans
, pending
, &to_reserve
);
945 btrfs_orphan_pre_snapshot(trans
, pending
, &to_reserve
);
947 if (to_reserve
> 0) {
948 ret
= btrfs_block_rsv_add(trans
, root
, &pending
->block_rsv
,
951 pending
->error
= ret
;
956 key
.objectid
= objectid
;
957 key
.offset
= (u64
)-1;
958 key
.type
= BTRFS_ROOT_ITEM_KEY
;
960 trans
->block_rsv
= &pending
->block_rsv
;
962 dentry
= pending
->dentry
;
963 parent
= dget_parent(dentry
);
964 parent_inode
= parent
->d_inode
;
965 parent_root
= BTRFS_I(parent_inode
)->root
;
966 record_root_in_trans(trans
, parent_root
);
969 * insert the directory item
971 ret
= btrfs_set_inode_index(parent_inode
, &index
);
973 ret
= btrfs_insert_dir_item(trans
, parent_root
,
974 dentry
->d_name
.name
, dentry
->d_name
.len
,
975 btrfs_ino(parent_inode
), &key
,
976 BTRFS_FT_DIR
, index
);
979 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
980 dentry
->d_name
.len
* 2);
981 ret
= btrfs_update_inode(trans
, parent_root
, parent_inode
);
984 record_root_in_trans(trans
, root
);
985 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
986 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
987 btrfs_check_and_init_root_item(new_root_item
);
989 root_flags
= btrfs_root_flags(new_root_item
);
990 if (pending
->readonly
)
991 root_flags
|= BTRFS_ROOT_SUBVOL_RDONLY
;
993 root_flags
&= ~BTRFS_ROOT_SUBVOL_RDONLY
;
994 btrfs_set_root_flags(new_root_item
, root_flags
);
996 old
= btrfs_lock_root_node(root
);
997 btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
);
998 btrfs_set_lock_blocking(old
);
1000 btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
1001 btrfs_tree_unlock(old
);
1002 free_extent_buffer(old
);
1004 btrfs_set_root_node(new_root_item
, tmp
);
1005 /* record when the snapshot was created in key.offset */
1006 key
.offset
= trans
->transid
;
1007 ret
= btrfs_insert_root(trans
, tree_root
, &key
, new_root_item
);
1008 btrfs_tree_unlock(tmp
);
1009 free_extent_buffer(tmp
);
1013 * insert root back/forward references
1015 ret
= btrfs_add_root_ref(trans
, tree_root
, objectid
,
1016 parent_root
->root_key
.objectid
,
1017 btrfs_ino(parent_inode
), index
,
1018 dentry
->d_name
.name
, dentry
->d_name
.len
);
1022 key
.offset
= (u64
)-1;
1023 pending
->snap
= btrfs_read_fs_root_no_name(root
->fs_info
, &key
);
1024 BUG_ON(IS_ERR(pending
->snap
));
1026 btrfs_reloc_post_snapshot(trans
, pending
);
1027 btrfs_orphan_post_snapshot(trans
, pending
);
1029 kfree(new_root_item
);
1030 btrfs_block_rsv_release(root
, &pending
->block_rsv
, (u64
)-1);
1035 * create all the snapshots we've scheduled for creation
1037 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
,
1038 struct btrfs_fs_info
*fs_info
)
1040 struct btrfs_pending_snapshot
*pending
;
1041 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
1044 list_for_each_entry(pending
, head
, list
) {
1045 ret
= create_pending_snapshot(trans
, fs_info
, pending
);
1051 static void update_super_roots(struct btrfs_root
*root
)
1053 struct btrfs_root_item
*root_item
;
1054 struct btrfs_super_block
*super
;
1056 super
= &root
->fs_info
->super_copy
;
1058 root_item
= &root
->fs_info
->chunk_root
->root_item
;
1059 super
->chunk_root
= root_item
->bytenr
;
1060 super
->chunk_root_generation
= root_item
->generation
;
1061 super
->chunk_root_level
= root_item
->level
;
1063 root_item
= &root
->fs_info
->tree_root
->root_item
;
1064 super
->root
= root_item
->bytenr
;
1065 super
->generation
= root_item
->generation
;
1066 super
->root_level
= root_item
->level
;
1067 if (super
->cache_generation
!= 0 || btrfs_test_opt(root
, SPACE_CACHE
))
1068 super
->cache_generation
= root_item
->generation
;
1071 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
1074 spin_lock(&info
->new_trans_lock
);
1075 if (info
->running_transaction
)
1076 ret
= info
->running_transaction
->in_commit
;
1077 spin_unlock(&info
->new_trans_lock
);
1081 int btrfs_transaction_blocked(struct btrfs_fs_info
*info
)
1084 spin_lock(&info
->new_trans_lock
);
1085 if (info
->running_transaction
)
1086 ret
= info
->running_transaction
->blocked
;
1087 spin_unlock(&info
->new_trans_lock
);
1092 * wait for the current transaction commit to start and block subsequent
1095 static void wait_current_trans_commit_start(struct btrfs_root
*root
,
1096 struct btrfs_transaction
*trans
)
1100 if (trans
->in_commit
)
1104 prepare_to_wait(&root
->fs_info
->transaction_blocked_wait
, &wait
,
1105 TASK_UNINTERRUPTIBLE
);
1106 if (trans
->in_commit
) {
1107 finish_wait(&root
->fs_info
->transaction_blocked_wait
,
1111 mutex_unlock(&root
->fs_info
->trans_mutex
);
1113 mutex_lock(&root
->fs_info
->trans_mutex
);
1114 finish_wait(&root
->fs_info
->transaction_blocked_wait
, &wait
);
1119 * wait for the current transaction to start and then become unblocked.
1122 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root
*root
,
1123 struct btrfs_transaction
*trans
)
1127 if (trans
->commit_done
|| (trans
->in_commit
&& !trans
->blocked
))
1131 prepare_to_wait(&root
->fs_info
->transaction_wait
, &wait
,
1132 TASK_UNINTERRUPTIBLE
);
1133 if (trans
->commit_done
||
1134 (trans
->in_commit
&& !trans
->blocked
)) {
1135 finish_wait(&root
->fs_info
->transaction_wait
,
1139 mutex_unlock(&root
->fs_info
->trans_mutex
);
1141 mutex_lock(&root
->fs_info
->trans_mutex
);
1142 finish_wait(&root
->fs_info
->transaction_wait
,
1148 * commit transactions asynchronously. once btrfs_commit_transaction_async
1149 * returns, any subsequent transaction will not be allowed to join.
1151 struct btrfs_async_commit
{
1152 struct btrfs_trans_handle
*newtrans
;
1153 struct btrfs_root
*root
;
1154 struct delayed_work work
;
1157 static void do_async_commit(struct work_struct
*work
)
1159 struct btrfs_async_commit
*ac
=
1160 container_of(work
, struct btrfs_async_commit
, work
.work
);
1162 btrfs_commit_transaction(ac
->newtrans
, ac
->root
);
1166 int btrfs_commit_transaction_async(struct btrfs_trans_handle
*trans
,
1167 struct btrfs_root
*root
,
1168 int wait_for_unblock
)
1170 struct btrfs_async_commit
*ac
;
1171 struct btrfs_transaction
*cur_trans
;
1173 ac
= kmalloc(sizeof(*ac
), GFP_NOFS
);
1177 INIT_DELAYED_WORK(&ac
->work
, do_async_commit
);
1179 ac
->newtrans
= btrfs_join_transaction(root
, 0);
1180 if (IS_ERR(ac
->newtrans
)) {
1181 int err
= PTR_ERR(ac
->newtrans
);
1186 /* take transaction reference */
1187 mutex_lock(&root
->fs_info
->trans_mutex
);
1188 cur_trans
= trans
->transaction
;
1189 atomic_inc(&cur_trans
->use_count
);
1190 mutex_unlock(&root
->fs_info
->trans_mutex
);
1192 btrfs_end_transaction(trans
, root
);
1193 schedule_delayed_work(&ac
->work
, 0);
1195 /* wait for transaction to start and unblock */
1196 mutex_lock(&root
->fs_info
->trans_mutex
);
1197 if (wait_for_unblock
)
1198 wait_current_trans_commit_start_and_unblock(root
, cur_trans
);
1200 wait_current_trans_commit_start(root
, cur_trans
);
1201 put_transaction(cur_trans
);
1202 mutex_unlock(&root
->fs_info
->trans_mutex
);
1208 * btrfs_transaction state sequence:
1209 * in_commit = 0, blocked = 0 (initial)
1210 * in_commit = 1, blocked = 1
1214 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
,
1215 struct btrfs_root
*root
)
1217 unsigned long joined
= 0;
1218 struct btrfs_transaction
*cur_trans
;
1219 struct btrfs_transaction
*prev_trans
= NULL
;
1222 int should_grow
= 0;
1223 unsigned long now
= get_seconds();
1224 int flush_on_commit
= btrfs_test_opt(root
, FLUSHONCOMMIT
);
1226 btrfs_run_ordered_operations(root
, 0);
1228 /* make a pass through all the delayed refs we have so far
1229 * any runnings procs may add more while we are here
1231 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1234 btrfs_trans_release_metadata(trans
, root
);
1236 cur_trans
= trans
->transaction
;
1238 * set the flushing flag so procs in this transaction have to
1239 * start sending their work down.
1241 cur_trans
->delayed_refs
.flushing
= 1;
1243 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1246 mutex_lock(&root
->fs_info
->trans_mutex
);
1247 if (cur_trans
->in_commit
) {
1248 atomic_inc(&cur_trans
->use_count
);
1249 mutex_unlock(&root
->fs_info
->trans_mutex
);
1250 btrfs_end_transaction(trans
, root
);
1252 ret
= wait_for_commit(root
, cur_trans
);
1255 mutex_lock(&root
->fs_info
->trans_mutex
);
1256 put_transaction(cur_trans
);
1257 mutex_unlock(&root
->fs_info
->trans_mutex
);
1262 trans
->transaction
->in_commit
= 1;
1263 trans
->transaction
->blocked
= 1;
1264 wake_up(&root
->fs_info
->transaction_blocked_wait
);
1266 if (cur_trans
->list
.prev
!= &root
->fs_info
->trans_list
) {
1267 prev_trans
= list_entry(cur_trans
->list
.prev
,
1268 struct btrfs_transaction
, list
);
1269 if (!prev_trans
->commit_done
) {
1270 atomic_inc(&prev_trans
->use_count
);
1271 mutex_unlock(&root
->fs_info
->trans_mutex
);
1273 wait_for_commit(root
, prev_trans
);
1275 mutex_lock(&root
->fs_info
->trans_mutex
);
1276 put_transaction(prev_trans
);
1280 if (now
< cur_trans
->start_time
|| now
- cur_trans
->start_time
< 1)
1284 int snap_pending
= 0;
1285 joined
= cur_trans
->num_joined
;
1286 if (!list_empty(&trans
->transaction
->pending_snapshots
))
1289 WARN_ON(cur_trans
!= trans
->transaction
);
1290 mutex_unlock(&root
->fs_info
->trans_mutex
);
1292 if (flush_on_commit
|| snap_pending
) {
1293 btrfs_start_delalloc_inodes(root
, 1);
1294 ret
= btrfs_wait_ordered_extents(root
, 0, 1);
1299 * rename don't use btrfs_join_transaction, so, once we
1300 * set the transaction to blocked above, we aren't going
1301 * to get any new ordered operations. We can safely run
1302 * it here and no for sure that nothing new will be added
1305 btrfs_run_ordered_operations(root
, 1);
1307 prepare_to_wait(&cur_trans
->writer_wait
, &wait
,
1308 TASK_UNINTERRUPTIBLE
);
1311 if (atomic_read(&cur_trans
->num_writers
) > 1)
1312 schedule_timeout(MAX_SCHEDULE_TIMEOUT
);
1313 else if (should_grow
)
1314 schedule_timeout(1);
1316 mutex_lock(&root
->fs_info
->trans_mutex
);
1317 finish_wait(&cur_trans
->writer_wait
, &wait
);
1318 } while (atomic_read(&cur_trans
->num_writers
) > 1 ||
1319 (should_grow
&& cur_trans
->num_joined
!= joined
));
1321 ret
= create_pending_snapshots(trans
, root
->fs_info
);
1324 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1327 WARN_ON(cur_trans
!= trans
->transaction
);
1329 /* btrfs_commit_tree_roots is responsible for getting the
1330 * various roots consistent with each other. Every pointer
1331 * in the tree of tree roots has to point to the most up to date
1332 * root for every subvolume and other tree. So, we have to keep
1333 * the tree logging code from jumping in and changing any
1336 * At this point in the commit, there can't be any tree-log
1337 * writers, but a little lower down we drop the trans mutex
1338 * and let new people in. By holding the tree_log_mutex
1339 * from now until after the super is written, we avoid races
1340 * with the tree-log code.
1342 mutex_lock(&root
->fs_info
->tree_log_mutex
);
1344 ret
= commit_fs_roots(trans
, root
);
1347 /* commit_fs_roots gets rid of all the tree log roots, it is now
1348 * safe to free the root of tree log roots
1350 btrfs_free_log_root_tree(trans
, root
->fs_info
);
1352 ret
= commit_cowonly_roots(trans
, root
);
1355 btrfs_prepare_extent_commit(trans
, root
);
1357 cur_trans
= root
->fs_info
->running_transaction
;
1358 spin_lock(&root
->fs_info
->new_trans_lock
);
1359 root
->fs_info
->running_transaction
= NULL
;
1360 spin_unlock(&root
->fs_info
->new_trans_lock
);
1362 btrfs_set_root_node(&root
->fs_info
->tree_root
->root_item
,
1363 root
->fs_info
->tree_root
->node
);
1364 switch_commit_root(root
->fs_info
->tree_root
);
1366 btrfs_set_root_node(&root
->fs_info
->chunk_root
->root_item
,
1367 root
->fs_info
->chunk_root
->node
);
1368 switch_commit_root(root
->fs_info
->chunk_root
);
1370 update_super_roots(root
);
1372 if (!root
->fs_info
->log_root_recovering
) {
1373 btrfs_set_super_log_root(&root
->fs_info
->super_copy
, 0);
1374 btrfs_set_super_log_root_level(&root
->fs_info
->super_copy
, 0);
1377 memcpy(&root
->fs_info
->super_for_commit
, &root
->fs_info
->super_copy
,
1378 sizeof(root
->fs_info
->super_copy
));
1380 trans
->transaction
->blocked
= 0;
1382 wake_up(&root
->fs_info
->transaction_wait
);
1384 mutex_unlock(&root
->fs_info
->trans_mutex
);
1385 ret
= btrfs_write_and_wait_transaction(trans
, root
);
1387 write_ctree_super(trans
, root
, 0);
1390 * the super is written, we can safely allow the tree-loggers
1391 * to go about their business
1393 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1395 btrfs_finish_extent_commit(trans
, root
);
1397 mutex_lock(&root
->fs_info
->trans_mutex
);
1399 cur_trans
->commit_done
= 1;
1401 root
->fs_info
->last_trans_committed
= cur_trans
->transid
;
1403 wake_up(&cur_trans
->commit_wait
);
1405 list_del_init(&cur_trans
->list
);
1406 put_transaction(cur_trans
);
1407 put_transaction(cur_trans
);
1409 trace_btrfs_transaction_commit(root
);
1411 mutex_unlock(&root
->fs_info
->trans_mutex
);
1413 if (current
->journal_info
== trans
)
1414 current
->journal_info
= NULL
;
1416 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1418 if (current
!= root
->fs_info
->transaction_kthread
)
1419 btrfs_run_delayed_iputs(root
);
1425 * interface function to delete all the snapshots we have scheduled for deletion
1427 int btrfs_clean_old_snapshots(struct btrfs_root
*root
)
1430 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1432 mutex_lock(&fs_info
->trans_mutex
);
1433 list_splice_init(&fs_info
->dead_roots
, &list
);
1434 mutex_unlock(&fs_info
->trans_mutex
);
1436 while (!list_empty(&list
)) {
1437 root
= list_entry(list
.next
, struct btrfs_root
, root_list
);
1438 list_del(&root
->root_list
);
1440 if (btrfs_header_backref_rev(root
->node
) <
1441 BTRFS_MIXED_BACKREF_REV
)
1442 btrfs_drop_snapshot(root
, NULL
, 0);
1444 btrfs_drop_snapshot(root
, NULL
, 1);