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(atomic_read(&transaction
->use_count
) == 0);
36 if (atomic_dec_and_test(&transaction
->use_count
)) {
37 BUG_ON(!list_empty(&transaction
->list
));
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
, int nofail
)
54 struct btrfs_transaction
*cur_trans
;
56 spin_lock(&root
->fs_info
->trans_lock
);
57 if (root
->fs_info
->trans_no_join
) {
59 spin_unlock(&root
->fs_info
->trans_lock
);
64 cur_trans
= root
->fs_info
->running_transaction
;
66 atomic_inc(&cur_trans
->use_count
);
67 atomic_inc(&cur_trans
->num_writers
);
68 cur_trans
->num_joined
++;
69 spin_unlock(&root
->fs_info
->trans_lock
);
72 spin_unlock(&root
->fs_info
->trans_lock
);
74 cur_trans
= kmem_cache_alloc(btrfs_transaction_cachep
, GFP_NOFS
);
77 spin_lock(&root
->fs_info
->trans_lock
);
78 if (root
->fs_info
->running_transaction
) {
79 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
80 cur_trans
= root
->fs_info
->running_transaction
;
81 atomic_inc(&cur_trans
->use_count
);
82 atomic_inc(&cur_trans
->num_writers
);
83 cur_trans
->num_joined
++;
84 spin_unlock(&root
->fs_info
->trans_lock
);
87 atomic_set(&cur_trans
->num_writers
, 1);
88 cur_trans
->num_joined
= 0;
89 init_waitqueue_head(&cur_trans
->writer_wait
);
90 init_waitqueue_head(&cur_trans
->commit_wait
);
91 cur_trans
->in_commit
= 0;
92 cur_trans
->blocked
= 0;
94 * One for this trans handle, one so it will live on until we
95 * commit the transaction.
97 atomic_set(&cur_trans
->use_count
, 2);
98 cur_trans
->commit_done
= 0;
99 cur_trans
->start_time
= get_seconds();
101 cur_trans
->delayed_refs
.root
= RB_ROOT
;
102 cur_trans
->delayed_refs
.num_entries
= 0;
103 cur_trans
->delayed_refs
.num_heads_ready
= 0;
104 cur_trans
->delayed_refs
.num_heads
= 0;
105 cur_trans
->delayed_refs
.flushing
= 0;
106 cur_trans
->delayed_refs
.run_delayed_start
= 0;
107 spin_lock_init(&cur_trans
->commit_lock
);
108 spin_lock_init(&cur_trans
->delayed_refs
.lock
);
110 INIT_LIST_HEAD(&cur_trans
->pending_snapshots
);
111 list_add_tail(&cur_trans
->list
, &root
->fs_info
->trans_list
);
112 extent_io_tree_init(&cur_trans
->dirty_pages
,
113 root
->fs_info
->btree_inode
->i_mapping
,
115 root
->fs_info
->generation
++;
116 cur_trans
->transid
= root
->fs_info
->generation
;
117 root
->fs_info
->running_transaction
= cur_trans
;
118 spin_unlock(&root
->fs_info
->trans_lock
);
124 * this does all the record keeping required to make sure that a reference
125 * counted root is properly recorded in a given transaction. This is required
126 * to make sure the old root from before we joined the transaction is deleted
127 * when the transaction commits
129 int btrfs_record_root_in_trans(struct btrfs_trans_handle
*trans
,
130 struct btrfs_root
*root
)
132 if (root
->ref_cows
&& root
->last_trans
< trans
->transid
) {
133 WARN_ON(root
== root
->fs_info
->extent_root
);
134 WARN_ON(root
->commit_root
!= root
->node
);
136 spin_lock(&root
->fs_info
->fs_roots_radix_lock
);
137 if (root
->last_trans
== trans
->transid
) {
138 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
141 root
->last_trans
= trans
->transid
;
142 radix_tree_tag_set(&root
->fs_info
->fs_roots_radix
,
143 (unsigned long)root
->root_key
.objectid
,
144 BTRFS_ROOT_TRANS_TAG
);
145 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
146 btrfs_init_reloc_root(trans
, root
);
151 /* wait for commit against the current transaction to become unblocked
152 * when this is done, it is safe to start a new transaction, but the current
153 * transaction might not be fully on disk.
155 static void wait_current_trans(struct btrfs_root
*root
)
157 struct btrfs_transaction
*cur_trans
;
159 spin_lock(&root
->fs_info
->trans_lock
);
160 cur_trans
= root
->fs_info
->running_transaction
;
161 if (cur_trans
&& cur_trans
->blocked
) {
163 atomic_inc(&cur_trans
->use_count
);
164 spin_unlock(&root
->fs_info
->trans_lock
);
166 prepare_to_wait(&root
->fs_info
->transaction_wait
, &wait
,
167 TASK_UNINTERRUPTIBLE
);
168 if (!cur_trans
->blocked
)
172 finish_wait(&root
->fs_info
->transaction_wait
, &wait
);
173 put_transaction(cur_trans
);
175 spin_unlock(&root
->fs_info
->trans_lock
);
179 enum btrfs_trans_type
{
186 static int may_wait_transaction(struct btrfs_root
*root
, int type
)
188 if (root
->fs_info
->log_root_recovering
)
191 if (type
== TRANS_USERSPACE
)
194 if (type
== TRANS_START
&&
195 !atomic_read(&root
->fs_info
->open_ioctl_trans
))
201 static struct btrfs_trans_handle
*start_transaction(struct btrfs_root
*root
,
202 u64 num_items
, int type
)
204 struct btrfs_trans_handle
*h
;
205 struct btrfs_transaction
*cur_trans
;
209 if (root
->fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
)
210 return ERR_PTR(-EROFS
);
212 if (current
->journal_info
) {
213 WARN_ON(type
!= TRANS_JOIN
&& type
!= TRANS_JOIN_NOLOCK
);
214 h
= current
->journal_info
;
216 h
->orig_rsv
= h
->block_rsv
;
221 h
= kmem_cache_alloc(btrfs_trans_handle_cachep
, GFP_NOFS
);
223 return ERR_PTR(-ENOMEM
);
225 if (may_wait_transaction(root
, type
))
226 wait_current_trans(root
);
229 ret
= join_transaction(root
, type
== TRANS_JOIN_NOLOCK
);
231 wait_current_trans(root
);
232 } while (ret
== -EBUSY
);
235 kmem_cache_free(btrfs_trans_handle_cachep
, h
);
239 cur_trans
= root
->fs_info
->running_transaction
;
241 h
->transid
= cur_trans
->transid
;
242 h
->transaction
= cur_trans
;
245 h
->bytes_reserved
= 0;
246 h
->delayed_ref_updates
= 0;
252 if (cur_trans
->blocked
&& may_wait_transaction(root
, type
)) {
253 btrfs_commit_transaction(h
, root
);
258 ret
= btrfs_trans_reserve_metadata(h
, root
, num_items
);
259 if (ret
== -EAGAIN
&& !retries
) {
261 btrfs_commit_transaction(h
, root
);
263 } else if (ret
== -EAGAIN
) {
265 * We have already retried and got EAGAIN, so really we
266 * don't have space, so set ret to -ENOSPC.
272 btrfs_end_transaction(h
, root
);
278 btrfs_record_root_in_trans(h
, root
);
280 if (!current
->journal_info
&& type
!= TRANS_USERSPACE
)
281 current
->journal_info
= h
;
285 struct btrfs_trans_handle
*btrfs_start_transaction(struct btrfs_root
*root
,
288 return start_transaction(root
, num_items
, TRANS_START
);
290 struct btrfs_trans_handle
*btrfs_join_transaction(struct btrfs_root
*root
)
292 return start_transaction(root
, 0, TRANS_JOIN
);
295 struct btrfs_trans_handle
*btrfs_join_transaction_nolock(struct btrfs_root
*root
)
297 return start_transaction(root
, 0, TRANS_JOIN_NOLOCK
);
300 struct btrfs_trans_handle
*btrfs_start_ioctl_transaction(struct btrfs_root
*root
)
302 return start_transaction(root
, 0, TRANS_USERSPACE
);
305 /* wait for a transaction commit to be fully complete */
306 static noinline
int wait_for_commit(struct btrfs_root
*root
,
307 struct btrfs_transaction
*commit
)
310 while (!commit
->commit_done
) {
311 prepare_to_wait(&commit
->commit_wait
, &wait
,
312 TASK_UNINTERRUPTIBLE
);
313 if (commit
->commit_done
)
317 finish_wait(&commit
->commit_wait
, &wait
);
321 int btrfs_wait_for_commit(struct btrfs_root
*root
, u64 transid
)
323 struct btrfs_transaction
*cur_trans
= NULL
, *t
;
328 if (transid
<= root
->fs_info
->last_trans_committed
)
331 /* find specified transaction */
332 spin_lock(&root
->fs_info
->trans_lock
);
333 list_for_each_entry(t
, &root
->fs_info
->trans_list
, list
) {
334 if (t
->transid
== transid
) {
336 atomic_inc(&cur_trans
->use_count
);
339 if (t
->transid
> transid
)
342 spin_unlock(&root
->fs_info
->trans_lock
);
345 goto out
; /* bad transid */
347 /* find newest transaction that is committing | committed */
348 spin_lock(&root
->fs_info
->trans_lock
);
349 list_for_each_entry_reverse(t
, &root
->fs_info
->trans_list
,
355 atomic_inc(&cur_trans
->use_count
);
359 spin_unlock(&root
->fs_info
->trans_lock
);
361 goto out
; /* nothing committing|committed */
364 wait_for_commit(root
, cur_trans
);
366 put_transaction(cur_trans
);
374 * rate limit against the drop_snapshot code. This helps to slow down new
375 * operations if the drop_snapshot code isn't able to keep up.
377 static void throttle_on_drops(struct btrfs_root
*root
)
379 struct btrfs_fs_info
*info
= root
->fs_info
;
380 int harder_count
= 0;
383 if (atomic_read(&info
->throttles
)) {
386 thr
= atomic_read(&info
->throttle_gen
);
389 prepare_to_wait(&info
->transaction_throttle
,
390 &wait
, TASK_UNINTERRUPTIBLE
);
391 if (!atomic_read(&info
->throttles
)) {
392 finish_wait(&info
->transaction_throttle
, &wait
);
396 finish_wait(&info
->transaction_throttle
, &wait
);
397 } while (thr
== atomic_read(&info
->throttle_gen
));
400 if (root
->fs_info
->total_ref_cache_size
> 1 * 1024 * 1024 &&
404 if (root
->fs_info
->total_ref_cache_size
> 5 * 1024 * 1024 &&
408 if (root
->fs_info
->total_ref_cache_size
> 10 * 1024 * 1024 &&
415 void btrfs_throttle(struct btrfs_root
*root
)
417 if (!atomic_read(&root
->fs_info
->open_ioctl_trans
))
418 wait_current_trans(root
);
421 static int should_end_transaction(struct btrfs_trans_handle
*trans
,
422 struct btrfs_root
*root
)
425 ret
= btrfs_block_rsv_check(trans
, root
,
426 &root
->fs_info
->global_block_rsv
, 0, 5);
430 int btrfs_should_end_transaction(struct btrfs_trans_handle
*trans
,
431 struct btrfs_root
*root
)
433 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
437 if (cur_trans
->blocked
|| cur_trans
->delayed_refs
.flushing
)
440 updates
= trans
->delayed_ref_updates
;
441 trans
->delayed_ref_updates
= 0;
443 btrfs_run_delayed_refs(trans
, root
, updates
);
445 return should_end_transaction(trans
, root
);
448 static int __btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
449 struct btrfs_root
*root
, int throttle
, int lock
)
451 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
452 struct btrfs_fs_info
*info
= root
->fs_info
;
455 if (--trans
->use_count
) {
456 trans
->block_rsv
= trans
->orig_rsv
;
461 unsigned long cur
= trans
->delayed_ref_updates
;
462 trans
->delayed_ref_updates
= 0;
464 trans
->transaction
->delayed_refs
.num_heads_ready
> 64) {
465 trans
->delayed_ref_updates
= 0;
468 * do a full flush if the transaction is trying
471 if (trans
->transaction
->delayed_refs
.flushing
)
473 btrfs_run_delayed_refs(trans
, root
, cur
);
480 btrfs_trans_release_metadata(trans
, root
);
482 if (lock
&& !atomic_read(&root
->fs_info
->open_ioctl_trans
) &&
483 should_end_transaction(trans
, root
)) {
484 trans
->transaction
->blocked
= 1;
488 if (lock
&& cur_trans
->blocked
&& !cur_trans
->in_commit
) {
490 return btrfs_commit_transaction(trans
, root
);
492 wake_up_process(info
->transaction_kthread
);
495 WARN_ON(cur_trans
!= info
->running_transaction
);
496 WARN_ON(atomic_read(&cur_trans
->num_writers
) < 1);
497 atomic_dec(&cur_trans
->num_writers
);
500 if (waitqueue_active(&cur_trans
->writer_wait
))
501 wake_up(&cur_trans
->writer_wait
);
502 put_transaction(cur_trans
);
504 if (current
->journal_info
== trans
)
505 current
->journal_info
= NULL
;
506 memset(trans
, 0, sizeof(*trans
));
507 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
510 btrfs_run_delayed_iputs(root
);
515 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
516 struct btrfs_root
*root
)
518 return __btrfs_end_transaction(trans
, root
, 0, 1);
521 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
,
522 struct btrfs_root
*root
)
524 return __btrfs_end_transaction(trans
, root
, 1, 1);
527 int btrfs_end_transaction_nolock(struct btrfs_trans_handle
*trans
,
528 struct btrfs_root
*root
)
530 return __btrfs_end_transaction(trans
, root
, 0, 0);
534 * when btree blocks are allocated, they have some corresponding bits set for
535 * them in one of two extent_io trees. This is used to make sure all of
536 * those extents are sent to disk but does not wait on them
538 int btrfs_write_marked_extents(struct btrfs_root
*root
,
539 struct extent_io_tree
*dirty_pages
, int mark
)
545 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
551 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
555 while (start
<= end
) {
558 index
= start
>> PAGE_CACHE_SHIFT
;
559 start
= (u64
)(index
+ 1) << PAGE_CACHE_SHIFT
;
560 page
= find_get_page(btree_inode
->i_mapping
, index
);
564 btree_lock_page_hook(page
);
565 if (!page
->mapping
) {
567 page_cache_release(page
);
571 if (PageWriteback(page
)) {
573 wait_on_page_writeback(page
);
576 page_cache_release(page
);
580 err
= write_one_page(page
, 0);
583 page_cache_release(page
);
592 * when btree blocks are allocated, they have some corresponding bits set for
593 * them in one of two extent_io trees. This is used to make sure all of
594 * those extents are on disk for transaction or log commit. We wait
595 * on all the pages and clear them from the dirty pages state tree
597 int btrfs_wait_marked_extents(struct btrfs_root
*root
,
598 struct extent_io_tree
*dirty_pages
, int mark
)
604 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
610 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
615 clear_extent_bits(dirty_pages
, start
, end
, mark
, GFP_NOFS
);
616 while (start
<= end
) {
617 index
= start
>> PAGE_CACHE_SHIFT
;
618 start
= (u64
)(index
+ 1) << PAGE_CACHE_SHIFT
;
619 page
= find_get_page(btree_inode
->i_mapping
, index
);
622 if (PageDirty(page
)) {
623 btree_lock_page_hook(page
);
624 wait_on_page_writeback(page
);
625 err
= write_one_page(page
, 0);
629 wait_on_page_writeback(page
);
630 page_cache_release(page
);
640 * when btree blocks are allocated, they have some corresponding bits set for
641 * them in one of two extent_io trees. This is used to make sure all of
642 * those extents are on disk for transaction or log commit
644 int btrfs_write_and_wait_marked_extents(struct btrfs_root
*root
,
645 struct extent_io_tree
*dirty_pages
, int mark
)
650 ret
= btrfs_write_marked_extents(root
, dirty_pages
, mark
);
651 ret2
= btrfs_wait_marked_extents(root
, dirty_pages
, mark
);
655 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
,
656 struct btrfs_root
*root
)
658 if (!trans
|| !trans
->transaction
) {
659 struct inode
*btree_inode
;
660 btree_inode
= root
->fs_info
->btree_inode
;
661 return filemap_write_and_wait(btree_inode
->i_mapping
);
663 return btrfs_write_and_wait_marked_extents(root
,
664 &trans
->transaction
->dirty_pages
,
669 * this is used to update the root pointer in the tree of tree roots.
671 * But, in the case of the extent allocation tree, updating the root
672 * pointer may allocate blocks which may change the root of the extent
675 * So, this loops and repeats and makes sure the cowonly root didn't
676 * change while the root pointer was being updated in the metadata.
678 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
679 struct btrfs_root
*root
)
684 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
686 old_root_used
= btrfs_root_used(&root
->root_item
);
687 btrfs_write_dirty_block_groups(trans
, root
);
690 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
691 if (old_root_bytenr
== root
->node
->start
&&
692 old_root_used
== btrfs_root_used(&root
->root_item
))
695 btrfs_set_root_node(&root
->root_item
, root
->node
);
696 ret
= btrfs_update_root(trans
, tree_root
,
701 old_root_used
= btrfs_root_used(&root
->root_item
);
702 ret
= btrfs_write_dirty_block_groups(trans
, root
);
706 if (root
!= root
->fs_info
->extent_root
)
707 switch_commit_root(root
);
713 * update all the cowonly tree roots on disk
715 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
,
716 struct btrfs_root
*root
)
718 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
719 struct list_head
*next
;
720 struct extent_buffer
*eb
;
723 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
726 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
727 btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
, 0, &eb
);
728 btrfs_tree_unlock(eb
);
729 free_extent_buffer(eb
);
731 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
734 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
735 next
= fs_info
->dirty_cowonly_roots
.next
;
737 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
739 update_cowonly_root(trans
, root
);
742 down_write(&fs_info
->extent_commit_sem
);
743 switch_commit_root(fs_info
->extent_root
);
744 up_write(&fs_info
->extent_commit_sem
);
750 * dead roots are old snapshots that need to be deleted. This allocates
751 * a dirty root struct and adds it into the list of dead roots that need to
754 int btrfs_add_dead_root(struct btrfs_root
*root
)
756 spin_lock(&root
->fs_info
->trans_lock
);
757 list_add(&root
->root_list
, &root
->fs_info
->dead_roots
);
758 spin_unlock(&root
->fs_info
->trans_lock
);
763 * update all the cowonly tree roots on disk
765 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
,
766 struct btrfs_root
*root
)
768 struct btrfs_root
*gang
[8];
769 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
774 spin_lock(&fs_info
->fs_roots_radix_lock
);
776 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
779 BTRFS_ROOT_TRANS_TAG
);
782 for (i
= 0; i
< ret
; i
++) {
784 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
785 (unsigned long)root
->root_key
.objectid
,
786 BTRFS_ROOT_TRANS_TAG
);
787 spin_unlock(&fs_info
->fs_roots_radix_lock
);
789 btrfs_free_log(trans
, root
);
790 btrfs_update_reloc_root(trans
, root
);
791 btrfs_orphan_commit_root(trans
, root
);
793 if (root
->commit_root
!= root
->node
) {
794 switch_commit_root(root
);
795 btrfs_set_root_node(&root
->root_item
,
799 err
= btrfs_update_root(trans
, fs_info
->tree_root
,
802 spin_lock(&fs_info
->fs_roots_radix_lock
);
807 spin_unlock(&fs_info
->fs_roots_radix_lock
);
812 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
813 * otherwise every leaf in the btree is read and defragged.
815 int btrfs_defrag_root(struct btrfs_root
*root
, int cacheonly
)
817 struct btrfs_fs_info
*info
= root
->fs_info
;
818 struct btrfs_trans_handle
*trans
;
822 if (xchg(&root
->defrag_running
, 1))
826 trans
= btrfs_start_transaction(root
, 0);
828 return PTR_ERR(trans
);
830 ret
= btrfs_defrag_leaves(trans
, root
, cacheonly
);
832 nr
= trans
->blocks_used
;
833 btrfs_end_transaction(trans
, root
);
834 btrfs_btree_balance_dirty(info
->tree_root
, nr
);
837 if (root
->fs_info
->closing
|| ret
!= -EAGAIN
)
840 root
->defrag_running
= 0;
846 * when dropping snapshots, we generate a ton of delayed refs, and it makes
847 * sense not to join the transaction while it is trying to flush the current
848 * queue of delayed refs out.
850 * This is used by the drop snapshot code only
852 static noinline
int wait_transaction_pre_flush(struct btrfs_fs_info
*info
)
856 mutex_lock(&info
->trans_mutex
);
857 while (info
->running_transaction
&&
858 info
->running_transaction
->delayed_refs
.flushing
) {
859 prepare_to_wait(&info
->transaction_wait
, &wait
,
860 TASK_UNINTERRUPTIBLE
);
861 mutex_unlock(&info
->trans_mutex
);
865 mutex_lock(&info
->trans_mutex
);
866 finish_wait(&info
->transaction_wait
, &wait
);
868 mutex_unlock(&info
->trans_mutex
);
873 * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
876 int btrfs_drop_dead_root(struct btrfs_root
*root
)
878 struct btrfs_trans_handle
*trans
;
879 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
885 * we don't want to jump in and create a bunch of
886 * delayed refs if the transaction is starting to close
888 wait_transaction_pre_flush(tree_root
->fs_info
);
889 trans
= btrfs_start_transaction(tree_root
, 1);
892 * we've joined a transaction, make sure it isn't
895 if (trans
->transaction
->delayed_refs
.flushing
) {
896 btrfs_end_transaction(trans
, tree_root
);
900 ret
= btrfs_drop_snapshot(trans
, root
);
904 ret
= btrfs_update_root(trans
, tree_root
,
910 nr
= trans
->blocks_used
;
911 ret
= btrfs_end_transaction(trans
, tree_root
);
914 btrfs_btree_balance_dirty(tree_root
, nr
);
919 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
922 nr
= trans
->blocks_used
;
923 ret
= btrfs_end_transaction(trans
, tree_root
);
926 free_extent_buffer(root
->node
);
927 free_extent_buffer(root
->commit_root
);
930 btrfs_btree_balance_dirty(tree_root
, nr
);
936 * new snapshots need to be created at a very specific time in the
937 * transaction commit. This does the actual creation
939 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
940 struct btrfs_fs_info
*fs_info
,
941 struct btrfs_pending_snapshot
*pending
)
943 struct btrfs_key key
;
944 struct btrfs_root_item
*new_root_item
;
945 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
946 struct btrfs_root
*root
= pending
->root
;
947 struct btrfs_root
*parent_root
;
948 struct inode
*parent_inode
;
949 struct dentry
*parent
;
950 struct dentry
*dentry
;
951 struct extent_buffer
*tmp
;
952 struct extent_buffer
*old
;
959 new_root_item
= kmalloc(sizeof(*new_root_item
), GFP_NOFS
);
960 if (!new_root_item
) {
961 pending
->error
= -ENOMEM
;
965 ret
= btrfs_find_free_objectid(trans
, tree_root
, 0, &objectid
);
967 pending
->error
= ret
;
971 btrfs_reloc_pre_snapshot(trans
, pending
, &to_reserve
);
972 btrfs_orphan_pre_snapshot(trans
, pending
, &to_reserve
);
974 if (to_reserve
> 0) {
975 ret
= btrfs_block_rsv_add(trans
, root
, &pending
->block_rsv
,
978 pending
->error
= ret
;
983 key
.objectid
= objectid
;
984 key
.offset
= (u64
)-1;
985 key
.type
= BTRFS_ROOT_ITEM_KEY
;
987 trans
->block_rsv
= &pending
->block_rsv
;
989 dentry
= pending
->dentry
;
990 parent
= dget_parent(dentry
);
991 parent_inode
= parent
->d_inode
;
992 parent_root
= BTRFS_I(parent_inode
)->root
;
993 btrfs_record_root_in_trans(trans
, parent_root
);
996 * insert the directory item
998 ret
= btrfs_set_inode_index(parent_inode
, &index
);
1000 ret
= btrfs_insert_dir_item(trans
, parent_root
,
1001 dentry
->d_name
.name
, dentry
->d_name
.len
,
1002 parent_inode
->i_ino
, &key
,
1003 BTRFS_FT_DIR
, index
);
1006 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
1007 dentry
->d_name
.len
* 2);
1008 ret
= btrfs_update_inode(trans
, parent_root
, parent_inode
);
1011 btrfs_record_root_in_trans(trans
, root
);
1012 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
1013 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
1014 btrfs_check_and_init_root_item(new_root_item
);
1016 root_flags
= btrfs_root_flags(new_root_item
);
1017 if (pending
->readonly
)
1018 root_flags
|= BTRFS_ROOT_SUBVOL_RDONLY
;
1020 root_flags
&= ~BTRFS_ROOT_SUBVOL_RDONLY
;
1021 btrfs_set_root_flags(new_root_item
, root_flags
);
1023 old
= btrfs_lock_root_node(root
);
1024 btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
);
1025 btrfs_set_lock_blocking(old
);
1027 btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
1028 btrfs_tree_unlock(old
);
1029 free_extent_buffer(old
);
1031 btrfs_set_root_node(new_root_item
, tmp
);
1032 /* record when the snapshot was created in key.offset */
1033 key
.offset
= trans
->transid
;
1034 ret
= btrfs_insert_root(trans
, tree_root
, &key
, new_root_item
);
1035 btrfs_tree_unlock(tmp
);
1036 free_extent_buffer(tmp
);
1040 * insert root back/forward references
1042 ret
= btrfs_add_root_ref(trans
, tree_root
, objectid
,
1043 parent_root
->root_key
.objectid
,
1044 parent_inode
->i_ino
, index
,
1045 dentry
->d_name
.name
, dentry
->d_name
.len
);
1049 key
.offset
= (u64
)-1;
1050 pending
->snap
= btrfs_read_fs_root_no_name(root
->fs_info
, &key
);
1051 BUG_ON(IS_ERR(pending
->snap
));
1053 btrfs_reloc_post_snapshot(trans
, pending
);
1054 btrfs_orphan_post_snapshot(trans
, pending
);
1056 kfree(new_root_item
);
1057 btrfs_block_rsv_release(root
, &pending
->block_rsv
, (u64
)-1);
1062 * create all the snapshots we've scheduled for creation
1064 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
,
1065 struct btrfs_fs_info
*fs_info
)
1067 struct btrfs_pending_snapshot
*pending
;
1068 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
1071 list_for_each_entry(pending
, head
, list
) {
1072 ret
= create_pending_snapshot(trans
, fs_info
, pending
);
1078 static void update_super_roots(struct btrfs_root
*root
)
1080 struct btrfs_root_item
*root_item
;
1081 struct btrfs_super_block
*super
;
1083 super
= &root
->fs_info
->super_copy
;
1085 root_item
= &root
->fs_info
->chunk_root
->root_item
;
1086 super
->chunk_root
= root_item
->bytenr
;
1087 super
->chunk_root_generation
= root_item
->generation
;
1088 super
->chunk_root_level
= root_item
->level
;
1090 root_item
= &root
->fs_info
->tree_root
->root_item
;
1091 super
->root
= root_item
->bytenr
;
1092 super
->generation
= root_item
->generation
;
1093 super
->root_level
= root_item
->level
;
1094 if (super
->cache_generation
!= 0 || btrfs_test_opt(root
, SPACE_CACHE
))
1095 super
->cache_generation
= root_item
->generation
;
1098 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
1101 spin_lock(&info
->trans_lock
);
1102 if (info
->running_transaction
)
1103 ret
= info
->running_transaction
->in_commit
;
1104 spin_unlock(&info
->trans_lock
);
1108 int btrfs_transaction_blocked(struct btrfs_fs_info
*info
)
1111 spin_lock(&info
->trans_lock
);
1112 if (info
->running_transaction
)
1113 ret
= info
->running_transaction
->blocked
;
1114 spin_unlock(&info
->trans_lock
);
1119 * wait for the current transaction commit to start and block subsequent
1122 static void wait_current_trans_commit_start(struct btrfs_root
*root
,
1123 struct btrfs_transaction
*trans
)
1127 if (trans
->in_commit
)
1131 prepare_to_wait(&root
->fs_info
->transaction_blocked_wait
, &wait
,
1132 TASK_UNINTERRUPTIBLE
);
1133 if (trans
->in_commit
) {
1134 finish_wait(&root
->fs_info
->transaction_blocked_wait
,
1139 finish_wait(&root
->fs_info
->transaction_blocked_wait
, &wait
);
1144 * wait for the current transaction to start and then become unblocked.
1147 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root
*root
,
1148 struct btrfs_transaction
*trans
)
1152 if (trans
->commit_done
|| (trans
->in_commit
&& !trans
->blocked
))
1156 prepare_to_wait(&root
->fs_info
->transaction_wait
, &wait
,
1157 TASK_UNINTERRUPTIBLE
);
1158 if (trans
->commit_done
||
1159 (trans
->in_commit
&& !trans
->blocked
)) {
1160 finish_wait(&root
->fs_info
->transaction_wait
,
1165 finish_wait(&root
->fs_info
->transaction_wait
,
1171 * commit transactions asynchronously. once btrfs_commit_transaction_async
1172 * returns, any subsequent transaction will not be allowed to join.
1174 struct btrfs_async_commit
{
1175 struct btrfs_trans_handle
*newtrans
;
1176 struct btrfs_root
*root
;
1177 struct delayed_work work
;
1180 static void do_async_commit(struct work_struct
*work
)
1182 struct btrfs_async_commit
*ac
=
1183 container_of(work
, struct btrfs_async_commit
, work
.work
);
1185 btrfs_commit_transaction(ac
->newtrans
, ac
->root
);
1189 int btrfs_commit_transaction_async(struct btrfs_trans_handle
*trans
,
1190 struct btrfs_root
*root
,
1191 int wait_for_unblock
)
1193 struct btrfs_async_commit
*ac
;
1194 struct btrfs_transaction
*cur_trans
;
1196 ac
= kmalloc(sizeof(*ac
), GFP_NOFS
);
1200 INIT_DELAYED_WORK(&ac
->work
, do_async_commit
);
1202 ac
->newtrans
= btrfs_join_transaction(root
);
1203 if (IS_ERR(ac
->newtrans
)) {
1204 int err
= PTR_ERR(ac
->newtrans
);
1209 /* take transaction reference */
1210 cur_trans
= trans
->transaction
;
1211 atomic_inc(&cur_trans
->use_count
);
1213 btrfs_end_transaction(trans
, root
);
1214 schedule_delayed_work(&ac
->work
, 0);
1216 /* wait for transaction to start and unblock */
1217 if (wait_for_unblock
)
1218 wait_current_trans_commit_start_and_unblock(root
, cur_trans
);
1220 wait_current_trans_commit_start(root
, cur_trans
);
1221 put_transaction(cur_trans
);
1227 * btrfs_transaction state sequence:
1228 * in_commit = 0, blocked = 0 (initial)
1229 * in_commit = 1, blocked = 1
1233 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
,
1234 struct btrfs_root
*root
)
1236 unsigned long joined
= 0;
1237 struct btrfs_transaction
*cur_trans
;
1238 struct btrfs_transaction
*prev_trans
= NULL
;
1241 int should_grow
= 0;
1242 unsigned long now
= get_seconds();
1243 int flush_on_commit
= btrfs_test_opt(root
, FLUSHONCOMMIT
);
1245 btrfs_run_ordered_operations(root
, 0);
1247 /* make a pass through all the delayed refs we have so far
1248 * any runnings procs may add more while we are here
1250 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1253 btrfs_trans_release_metadata(trans
, root
);
1255 cur_trans
= trans
->transaction
;
1257 * set the flushing flag so procs in this transaction have to
1258 * start sending their work down.
1260 cur_trans
->delayed_refs
.flushing
= 1;
1262 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1265 spin_lock(&cur_trans
->commit_lock
);
1266 if (cur_trans
->in_commit
) {
1267 spin_unlock(&cur_trans
->commit_lock
);
1268 atomic_inc(&cur_trans
->use_count
);
1269 btrfs_end_transaction(trans
, root
);
1271 ret
= wait_for_commit(root
, cur_trans
);
1274 put_transaction(cur_trans
);
1279 trans
->transaction
->in_commit
= 1;
1280 trans
->transaction
->blocked
= 1;
1281 spin_unlock(&cur_trans
->commit_lock
);
1282 wake_up(&root
->fs_info
->transaction_blocked_wait
);
1284 spin_lock(&root
->fs_info
->trans_lock
);
1285 if (cur_trans
->list
.prev
!= &root
->fs_info
->trans_list
) {
1286 prev_trans
= list_entry(cur_trans
->list
.prev
,
1287 struct btrfs_transaction
, list
);
1288 if (!prev_trans
->commit_done
) {
1289 atomic_inc(&prev_trans
->use_count
);
1290 spin_unlock(&root
->fs_info
->trans_lock
);
1292 wait_for_commit(root
, prev_trans
);
1294 put_transaction(prev_trans
);
1296 spin_unlock(&root
->fs_info
->trans_lock
);
1299 spin_unlock(&root
->fs_info
->trans_lock
);
1302 if (now
< cur_trans
->start_time
|| now
- cur_trans
->start_time
< 1)
1306 int snap_pending
= 0;
1308 joined
= cur_trans
->num_joined
;
1309 if (!list_empty(&trans
->transaction
->pending_snapshots
))
1312 WARN_ON(cur_trans
!= trans
->transaction
);
1314 if (flush_on_commit
|| snap_pending
) {
1315 btrfs_start_delalloc_inodes(root
, 1);
1316 ret
= btrfs_wait_ordered_extents(root
, 0, 1);
1321 * rename don't use btrfs_join_transaction, so, once we
1322 * set the transaction to blocked above, we aren't going
1323 * to get any new ordered operations. We can safely run
1324 * it here and no for sure that nothing new will be added
1327 btrfs_run_ordered_operations(root
, 1);
1329 prepare_to_wait(&cur_trans
->writer_wait
, &wait
,
1330 TASK_UNINTERRUPTIBLE
);
1332 if (atomic_read(&cur_trans
->num_writers
) > 1)
1333 schedule_timeout(MAX_SCHEDULE_TIMEOUT
);
1334 else if (should_grow
)
1335 schedule_timeout(1);
1337 finish_wait(&cur_trans
->writer_wait
, &wait
);
1338 spin_lock(&root
->fs_info
->trans_lock
);
1339 root
->fs_info
->trans_no_join
= 1;
1340 spin_unlock(&root
->fs_info
->trans_lock
);
1341 } while (atomic_read(&cur_trans
->num_writers
) > 1 ||
1342 (should_grow
&& cur_trans
->num_joined
!= joined
));
1344 ret
= create_pending_snapshots(trans
, root
->fs_info
);
1347 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1350 WARN_ON(cur_trans
!= trans
->transaction
);
1352 /* btrfs_commit_tree_roots is responsible for getting the
1353 * various roots consistent with each other. Every pointer
1354 * in the tree of tree roots has to point to the most up to date
1355 * root for every subvolume and other tree. So, we have to keep
1356 * the tree logging code from jumping in and changing any
1359 * At this point in the commit, there can't be any tree-log
1360 * writers, but a little lower down we drop the trans mutex
1361 * and let new people in. By holding the tree_log_mutex
1362 * from now until after the super is written, we avoid races
1363 * with the tree-log code.
1365 mutex_lock(&root
->fs_info
->tree_log_mutex
);
1367 ret
= commit_fs_roots(trans
, root
);
1370 /* commit_fs_roots gets rid of all the tree log roots, it is now
1371 * safe to free the root of tree log roots
1373 btrfs_free_log_root_tree(trans
, root
->fs_info
);
1375 ret
= commit_cowonly_roots(trans
, root
);
1378 btrfs_prepare_extent_commit(trans
, root
);
1380 cur_trans
= root
->fs_info
->running_transaction
;
1382 btrfs_set_root_node(&root
->fs_info
->tree_root
->root_item
,
1383 root
->fs_info
->tree_root
->node
);
1384 switch_commit_root(root
->fs_info
->tree_root
);
1386 btrfs_set_root_node(&root
->fs_info
->chunk_root
->root_item
,
1387 root
->fs_info
->chunk_root
->node
);
1388 switch_commit_root(root
->fs_info
->chunk_root
);
1390 update_super_roots(root
);
1392 if (!root
->fs_info
->log_root_recovering
) {
1393 btrfs_set_super_log_root(&root
->fs_info
->super_copy
, 0);
1394 btrfs_set_super_log_root_level(&root
->fs_info
->super_copy
, 0);
1397 memcpy(&root
->fs_info
->super_for_commit
, &root
->fs_info
->super_copy
,
1398 sizeof(root
->fs_info
->super_copy
));
1400 trans
->transaction
->blocked
= 0;
1401 spin_lock(&root
->fs_info
->trans_lock
);
1402 root
->fs_info
->running_transaction
= NULL
;
1403 root
->fs_info
->trans_no_join
= 0;
1404 spin_unlock(&root
->fs_info
->trans_lock
);
1406 wake_up(&root
->fs_info
->transaction_wait
);
1408 ret
= btrfs_write_and_wait_transaction(trans
, root
);
1410 write_ctree_super(trans
, root
, 0);
1413 * the super is written, we can safely allow the tree-loggers
1414 * to go about their business
1416 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1418 btrfs_finish_extent_commit(trans
, root
);
1420 cur_trans
->commit_done
= 1;
1422 root
->fs_info
->last_trans_committed
= cur_trans
->transid
;
1424 wake_up(&cur_trans
->commit_wait
);
1426 spin_lock(&root
->fs_info
->trans_lock
);
1427 list_del_init(&cur_trans
->list
);
1428 spin_unlock(&root
->fs_info
->trans_lock
);
1430 put_transaction(cur_trans
);
1431 put_transaction(cur_trans
);
1433 trace_btrfs_transaction_commit(root
);
1435 if (current
->journal_info
== trans
)
1436 current
->journal_info
= NULL
;
1438 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1440 if (current
!= root
->fs_info
->transaction_kthread
)
1441 btrfs_run_delayed_iputs(root
);
1447 * interface function to delete all the snapshots we have scheduled for deletion
1449 int btrfs_clean_old_snapshots(struct btrfs_root
*root
)
1452 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1454 spin_lock(&fs_info
->trans_lock
);
1455 list_splice_init(&fs_info
->dead_roots
, &list
);
1456 spin_unlock(&fs_info
->trans_lock
);
1458 while (!list_empty(&list
)) {
1459 root
= list_entry(list
.next
, struct btrfs_root
, root_list
);
1460 list_del(&root
->root_list
);
1462 if (btrfs_header_backref_rev(root
->node
) <
1463 BTRFS_MIXED_BACKREF_REV
)
1464 btrfs_drop_snapshot(root
, NULL
, 0);
1466 btrfs_drop_snapshot(root
, NULL
, 1);