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>
25 #include <linux/uuid.h>
28 #include "transaction.h"
31 #include "inode-map.h"
34 #define BTRFS_ROOT_TRANS_TAG 0
36 void put_transaction(struct btrfs_transaction
*transaction
)
38 WARN_ON(atomic_read(&transaction
->use_count
) == 0);
39 if (atomic_dec_and_test(&transaction
->use_count
)) {
40 BUG_ON(!list_empty(&transaction
->list
));
41 WARN_ON(transaction
->delayed_refs
.root
.rb_node
);
42 memset(transaction
, 0, sizeof(*transaction
));
43 kmem_cache_free(btrfs_transaction_cachep
, transaction
);
47 static noinline
void switch_commit_root(struct btrfs_root
*root
)
49 free_extent_buffer(root
->commit_root
);
50 root
->commit_root
= btrfs_root_node(root
);
54 * either allocate a new transaction or hop into the existing one
56 static noinline
int join_transaction(struct btrfs_root
*root
, int type
)
58 struct btrfs_transaction
*cur_trans
;
59 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
61 spin_lock(&fs_info
->trans_lock
);
63 /* The file system has been taken offline. No new transactions. */
64 if (fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
) {
65 spin_unlock(&fs_info
->trans_lock
);
69 if (fs_info
->trans_no_join
) {
71 * If we are JOIN_NOLOCK we're already committing a current
72 * transaction, we just need a handle to deal with something
73 * when committing the transaction, such as inode cache and
74 * space cache. It is a special case.
76 if (type
!= TRANS_JOIN_NOLOCK
) {
77 spin_unlock(&fs_info
->trans_lock
);
82 cur_trans
= fs_info
->running_transaction
;
84 if (cur_trans
->aborted
) {
85 spin_unlock(&fs_info
->trans_lock
);
86 return cur_trans
->aborted
;
88 atomic_inc(&cur_trans
->use_count
);
89 atomic_inc(&cur_trans
->num_writers
);
90 cur_trans
->num_joined
++;
91 spin_unlock(&fs_info
->trans_lock
);
94 spin_unlock(&fs_info
->trans_lock
);
97 * If we are ATTACH, we just want to catch the current transaction,
98 * and commit it. If there is no transaction, just return ENOENT.
100 if (type
== TRANS_ATTACH
)
103 cur_trans
= kmem_cache_alloc(btrfs_transaction_cachep
, GFP_NOFS
);
107 spin_lock(&fs_info
->trans_lock
);
108 if (fs_info
->running_transaction
) {
110 * someone started a transaction after we unlocked. Make sure
111 * to redo the trans_no_join checks above
113 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
114 cur_trans
= fs_info
->running_transaction
;
116 } else if (fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
) {
117 spin_unlock(&fs_info
->trans_lock
);
118 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
122 atomic_set(&cur_trans
->num_writers
, 1);
123 cur_trans
->num_joined
= 0;
124 init_waitqueue_head(&cur_trans
->writer_wait
);
125 init_waitqueue_head(&cur_trans
->commit_wait
);
126 cur_trans
->in_commit
= 0;
127 cur_trans
->blocked
= 0;
129 * One for this trans handle, one so it will live on until we
130 * commit the transaction.
132 atomic_set(&cur_trans
->use_count
, 2);
133 cur_trans
->commit_done
= 0;
134 cur_trans
->start_time
= get_seconds();
136 cur_trans
->delayed_refs
.root
= RB_ROOT
;
137 cur_trans
->delayed_refs
.num_entries
= 0;
138 cur_trans
->delayed_refs
.num_heads_ready
= 0;
139 cur_trans
->delayed_refs
.num_heads
= 0;
140 cur_trans
->delayed_refs
.flushing
= 0;
141 cur_trans
->delayed_refs
.run_delayed_start
= 0;
144 * although the tree mod log is per file system and not per transaction,
145 * the log must never go across transaction boundaries.
148 if (!list_empty(&fs_info
->tree_mod_seq_list
)) {
149 printk(KERN_ERR
"btrfs: tree_mod_seq_list not empty when "
150 "creating a fresh transaction\n");
153 if (!RB_EMPTY_ROOT(&fs_info
->tree_mod_log
)) {
154 printk(KERN_ERR
"btrfs: tree_mod_log rb tree not empty when "
155 "creating a fresh transaction\n");
158 atomic_set(&fs_info
->tree_mod_seq
, 0);
160 spin_lock_init(&cur_trans
->commit_lock
);
161 spin_lock_init(&cur_trans
->delayed_refs
.lock
);
163 INIT_LIST_HEAD(&cur_trans
->pending_snapshots
);
164 list_add_tail(&cur_trans
->list
, &fs_info
->trans_list
);
165 extent_io_tree_init(&cur_trans
->dirty_pages
,
166 fs_info
->btree_inode
->i_mapping
);
167 fs_info
->generation
++;
168 cur_trans
->transid
= fs_info
->generation
;
169 fs_info
->running_transaction
= cur_trans
;
170 cur_trans
->aborted
= 0;
171 spin_unlock(&fs_info
->trans_lock
);
177 * this does all the record keeping required to make sure that a reference
178 * counted root is properly recorded in a given transaction. This is required
179 * to make sure the old root from before we joined the transaction is deleted
180 * when the transaction commits
182 static int record_root_in_trans(struct btrfs_trans_handle
*trans
,
183 struct btrfs_root
*root
)
185 if (root
->ref_cows
&& root
->last_trans
< trans
->transid
) {
186 WARN_ON(root
== root
->fs_info
->extent_root
);
187 WARN_ON(root
->commit_root
!= root
->node
);
190 * see below for in_trans_setup usage rules
191 * we have the reloc mutex held now, so there
192 * is only one writer in this function
194 root
->in_trans_setup
= 1;
196 /* make sure readers find in_trans_setup before
197 * they find our root->last_trans update
201 spin_lock(&root
->fs_info
->fs_roots_radix_lock
);
202 if (root
->last_trans
== trans
->transid
) {
203 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
206 radix_tree_tag_set(&root
->fs_info
->fs_roots_radix
,
207 (unsigned long)root
->root_key
.objectid
,
208 BTRFS_ROOT_TRANS_TAG
);
209 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
210 root
->last_trans
= trans
->transid
;
212 /* this is pretty tricky. We don't want to
213 * take the relocation lock in btrfs_record_root_in_trans
214 * unless we're really doing the first setup for this root in
217 * Normally we'd use root->last_trans as a flag to decide
218 * if we want to take the expensive mutex.
220 * But, we have to set root->last_trans before we
221 * init the relocation root, otherwise, we trip over warnings
222 * in ctree.c. The solution used here is to flag ourselves
223 * with root->in_trans_setup. When this is 1, we're still
224 * fixing up the reloc trees and everyone must wait.
226 * When this is zero, they can trust root->last_trans and fly
227 * through btrfs_record_root_in_trans without having to take the
228 * lock. smp_wmb() makes sure that all the writes above are
229 * done before we pop in the zero below
231 btrfs_init_reloc_root(trans
, root
);
233 root
->in_trans_setup
= 0;
239 int btrfs_record_root_in_trans(struct btrfs_trans_handle
*trans
,
240 struct btrfs_root
*root
)
246 * see record_root_in_trans for comments about in_trans_setup usage
250 if (root
->last_trans
== trans
->transid
&&
251 !root
->in_trans_setup
)
254 mutex_lock(&root
->fs_info
->reloc_mutex
);
255 record_root_in_trans(trans
, root
);
256 mutex_unlock(&root
->fs_info
->reloc_mutex
);
261 /* wait for commit against the current transaction to become unblocked
262 * when this is done, it is safe to start a new transaction, but the current
263 * transaction might not be fully on disk.
265 static void wait_current_trans(struct btrfs_root
*root
)
267 struct btrfs_transaction
*cur_trans
;
269 spin_lock(&root
->fs_info
->trans_lock
);
270 cur_trans
= root
->fs_info
->running_transaction
;
271 if (cur_trans
&& cur_trans
->blocked
) {
272 atomic_inc(&cur_trans
->use_count
);
273 spin_unlock(&root
->fs_info
->trans_lock
);
275 wait_event(root
->fs_info
->transaction_wait
,
276 !cur_trans
->blocked
);
277 put_transaction(cur_trans
);
279 spin_unlock(&root
->fs_info
->trans_lock
);
283 static int may_wait_transaction(struct btrfs_root
*root
, int type
)
285 if (root
->fs_info
->log_root_recovering
)
288 if (type
== TRANS_USERSPACE
)
291 if (type
== TRANS_START
&&
292 !atomic_read(&root
->fs_info
->open_ioctl_trans
))
298 static struct btrfs_trans_handle
*
299 start_transaction(struct btrfs_root
*root
, u64 num_items
, int type
,
300 enum btrfs_reserve_flush_enum flush
)
302 struct btrfs_trans_handle
*h
;
303 struct btrfs_transaction
*cur_trans
;
306 u64 qgroup_reserved
= 0;
308 if (root
->fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
)
309 return ERR_PTR(-EROFS
);
311 if (current
->journal_info
) {
312 WARN_ON(type
!= TRANS_JOIN
&& type
!= TRANS_JOIN_NOLOCK
);
313 h
= current
->journal_info
;
315 h
->orig_rsv
= h
->block_rsv
;
321 * Do the reservation before we join the transaction so we can do all
322 * the appropriate flushing if need be.
324 if (num_items
> 0 && root
!= root
->fs_info
->chunk_root
) {
325 if (root
->fs_info
->quota_enabled
&&
326 is_fstree(root
->root_key
.objectid
)) {
327 qgroup_reserved
= num_items
* root
->leafsize
;
328 ret
= btrfs_qgroup_reserve(root
, qgroup_reserved
);
333 num_bytes
= btrfs_calc_trans_metadata_size(root
, num_items
);
334 ret
= btrfs_block_rsv_add(root
,
335 &root
->fs_info
->trans_block_rsv
,
341 h
= kmem_cache_alloc(btrfs_trans_handle_cachep
, GFP_NOFS
);
343 return ERR_PTR(-ENOMEM
);
346 * If we are JOIN_NOLOCK we're already committing a transaction and
347 * waiting on this guy, so we don't need to do the sb_start_intwrite
348 * because we're already holding a ref. We need this because we could
349 * have raced in and did an fsync() on a file which can kick a commit
350 * and then we deadlock with somebody doing a freeze.
352 * If we are ATTACH, it means we just want to catch the current
353 * transaction and commit it, so we needn't do sb_start_intwrite().
355 if (type
< TRANS_JOIN_NOLOCK
)
356 sb_start_intwrite(root
->fs_info
->sb
);
358 if (may_wait_transaction(root
, type
))
359 wait_current_trans(root
);
362 ret
= join_transaction(root
, type
);
364 wait_current_trans(root
);
365 } while (ret
== -EBUSY
);
368 /* We must get the transaction if we are JOIN_NOLOCK. */
369 BUG_ON(type
== TRANS_JOIN_NOLOCK
);
371 if (type
< TRANS_JOIN_NOLOCK
)
372 sb_end_intwrite(root
->fs_info
->sb
);
373 kmem_cache_free(btrfs_trans_handle_cachep
, h
);
377 cur_trans
= root
->fs_info
->running_transaction
;
379 h
->transid
= cur_trans
->transid
;
380 h
->transaction
= cur_trans
;
382 h
->bytes_reserved
= 0;
384 h
->delayed_ref_updates
= 0;
390 h
->qgroup_reserved
= qgroup_reserved
;
391 h
->delayed_ref_elem
.seq
= 0;
393 INIT_LIST_HEAD(&h
->qgroup_ref_list
);
394 INIT_LIST_HEAD(&h
->new_bgs
);
397 if (cur_trans
->blocked
&& may_wait_transaction(root
, type
)) {
398 btrfs_commit_transaction(h
, root
);
403 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
404 h
->transid
, num_bytes
, 1);
405 h
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
406 h
->bytes_reserved
= num_bytes
;
410 btrfs_record_root_in_trans(h
, root
);
412 if (!current
->journal_info
&& type
!= TRANS_USERSPACE
)
413 current
->journal_info
= h
;
417 struct btrfs_trans_handle
*btrfs_start_transaction(struct btrfs_root
*root
,
420 return start_transaction(root
, num_items
, TRANS_START
,
421 BTRFS_RESERVE_FLUSH_ALL
);
424 struct btrfs_trans_handle
*btrfs_start_transaction_lflush(
425 struct btrfs_root
*root
, int num_items
)
427 return start_transaction(root
, num_items
, TRANS_START
,
428 BTRFS_RESERVE_FLUSH_LIMIT
);
431 struct btrfs_trans_handle
*btrfs_join_transaction(struct btrfs_root
*root
)
433 return start_transaction(root
, 0, TRANS_JOIN
, 0);
436 struct btrfs_trans_handle
*btrfs_join_transaction_nolock(struct btrfs_root
*root
)
438 return start_transaction(root
, 0, TRANS_JOIN_NOLOCK
, 0);
441 struct btrfs_trans_handle
*btrfs_start_ioctl_transaction(struct btrfs_root
*root
)
443 return start_transaction(root
, 0, TRANS_USERSPACE
, 0);
446 struct btrfs_trans_handle
*btrfs_attach_transaction(struct btrfs_root
*root
)
448 return start_transaction(root
, 0, TRANS_ATTACH
, 0);
451 /* wait for a transaction commit to be fully complete */
452 static noinline
void wait_for_commit(struct btrfs_root
*root
,
453 struct btrfs_transaction
*commit
)
455 wait_event(commit
->commit_wait
, commit
->commit_done
);
458 int btrfs_wait_for_commit(struct btrfs_root
*root
, u64 transid
)
460 struct btrfs_transaction
*cur_trans
= NULL
, *t
;
465 if (transid
<= root
->fs_info
->last_trans_committed
)
468 /* find specified transaction */
469 spin_lock(&root
->fs_info
->trans_lock
);
470 list_for_each_entry(t
, &root
->fs_info
->trans_list
, list
) {
471 if (t
->transid
== transid
) {
473 atomic_inc(&cur_trans
->use_count
);
476 if (t
->transid
> transid
)
479 spin_unlock(&root
->fs_info
->trans_lock
);
482 goto out
; /* bad transid */
484 /* find newest transaction that is committing | committed */
485 spin_lock(&root
->fs_info
->trans_lock
);
486 list_for_each_entry_reverse(t
, &root
->fs_info
->trans_list
,
492 atomic_inc(&cur_trans
->use_count
);
496 spin_unlock(&root
->fs_info
->trans_lock
);
498 goto out
; /* nothing committing|committed */
501 wait_for_commit(root
, cur_trans
);
503 put_transaction(cur_trans
);
509 void btrfs_throttle(struct btrfs_root
*root
)
511 if (!atomic_read(&root
->fs_info
->open_ioctl_trans
))
512 wait_current_trans(root
);
515 static int should_end_transaction(struct btrfs_trans_handle
*trans
,
516 struct btrfs_root
*root
)
520 ret
= btrfs_block_rsv_check(root
, &root
->fs_info
->global_block_rsv
, 5);
524 int btrfs_should_end_transaction(struct btrfs_trans_handle
*trans
,
525 struct btrfs_root
*root
)
527 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
532 if (cur_trans
->blocked
|| cur_trans
->delayed_refs
.flushing
)
535 updates
= trans
->delayed_ref_updates
;
536 trans
->delayed_ref_updates
= 0;
538 err
= btrfs_run_delayed_refs(trans
, root
, updates
);
539 if (err
) /* Error code will also eval true */
543 return should_end_transaction(trans
, root
);
546 static int __btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
547 struct btrfs_root
*root
, int throttle
)
549 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
550 struct btrfs_fs_info
*info
= root
->fs_info
;
552 int lock
= (trans
->type
!= TRANS_JOIN_NOLOCK
);
555 if (--trans
->use_count
) {
556 trans
->block_rsv
= trans
->orig_rsv
;
561 * do the qgroup accounting as early as possible
563 err
= btrfs_delayed_refs_qgroup_accounting(trans
, info
);
565 btrfs_trans_release_metadata(trans
, root
);
566 trans
->block_rsv
= NULL
;
568 * the same root has to be passed to start_transaction and
569 * end_transaction. Subvolume quota depends on this.
571 WARN_ON(trans
->root
!= root
);
573 if (trans
->qgroup_reserved
) {
574 btrfs_qgroup_free(root
, trans
->qgroup_reserved
);
575 trans
->qgroup_reserved
= 0;
578 if (!list_empty(&trans
->new_bgs
))
579 btrfs_create_pending_block_groups(trans
, root
);
582 unsigned long cur
= trans
->delayed_ref_updates
;
583 trans
->delayed_ref_updates
= 0;
585 trans
->transaction
->delayed_refs
.num_heads_ready
> 64) {
586 trans
->delayed_ref_updates
= 0;
587 btrfs_run_delayed_refs(trans
, root
, cur
);
593 btrfs_trans_release_metadata(trans
, root
);
594 trans
->block_rsv
= NULL
;
596 if (!list_empty(&trans
->new_bgs
))
597 btrfs_create_pending_block_groups(trans
, root
);
599 if (lock
&& !atomic_read(&root
->fs_info
->open_ioctl_trans
) &&
600 should_end_transaction(trans
, root
)) {
601 trans
->transaction
->blocked
= 1;
605 if (lock
&& cur_trans
->blocked
&& !cur_trans
->in_commit
) {
608 * We may race with somebody else here so end up having
609 * to call end_transaction on ourselves again, so inc
613 return btrfs_commit_transaction(trans
, root
);
615 wake_up_process(info
->transaction_kthread
);
619 if (trans
->type
< TRANS_JOIN_NOLOCK
)
620 sb_end_intwrite(root
->fs_info
->sb
);
622 WARN_ON(cur_trans
!= info
->running_transaction
);
623 WARN_ON(atomic_read(&cur_trans
->num_writers
) < 1);
624 atomic_dec(&cur_trans
->num_writers
);
627 if (waitqueue_active(&cur_trans
->writer_wait
))
628 wake_up(&cur_trans
->writer_wait
);
629 put_transaction(cur_trans
);
631 if (current
->journal_info
== trans
)
632 current
->journal_info
= NULL
;
635 btrfs_run_delayed_iputs(root
);
637 if (trans
->aborted
||
638 root
->fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
) {
641 assert_qgroups_uptodate(trans
);
643 memset(trans
, 0, sizeof(*trans
));
644 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
648 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
649 struct btrfs_root
*root
)
653 ret
= __btrfs_end_transaction(trans
, root
, 0);
659 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
,
660 struct btrfs_root
*root
)
664 ret
= __btrfs_end_transaction(trans
, root
, 1);
670 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle
*trans
,
671 struct btrfs_root
*root
)
673 return __btrfs_end_transaction(trans
, root
, 1);
677 * when btree blocks are allocated, they have some corresponding bits set for
678 * them in one of two extent_io trees. This is used to make sure all of
679 * those extents are sent to disk but does not wait on them
681 int btrfs_write_marked_extents(struct btrfs_root
*root
,
682 struct extent_io_tree
*dirty_pages
, int mark
)
686 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
687 struct extent_state
*cached_state
= NULL
;
691 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
692 mark
, &cached_state
)) {
693 convert_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
,
694 mark
, &cached_state
, GFP_NOFS
);
696 err
= filemap_fdatawrite_range(mapping
, start
, end
);
708 * when btree blocks are allocated, they have some corresponding bits set for
709 * them in one of two extent_io trees. This is used to make sure all of
710 * those extents are on disk for transaction or log commit. We wait
711 * on all the pages and clear them from the dirty pages state tree
713 int btrfs_wait_marked_extents(struct btrfs_root
*root
,
714 struct extent_io_tree
*dirty_pages
, int mark
)
718 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
719 struct extent_state
*cached_state
= NULL
;
723 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
724 EXTENT_NEED_WAIT
, &cached_state
)) {
725 clear_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
,
726 0, 0, &cached_state
, GFP_NOFS
);
727 err
= filemap_fdatawait_range(mapping
, start
, end
);
739 * when btree blocks are allocated, they have some corresponding bits set for
740 * them in one of two extent_io trees. This is used to make sure all of
741 * those extents are on disk for transaction or log commit
743 int btrfs_write_and_wait_marked_extents(struct btrfs_root
*root
,
744 struct extent_io_tree
*dirty_pages
, int mark
)
749 ret
= btrfs_write_marked_extents(root
, dirty_pages
, mark
);
750 ret2
= btrfs_wait_marked_extents(root
, dirty_pages
, mark
);
759 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
,
760 struct btrfs_root
*root
)
762 if (!trans
|| !trans
->transaction
) {
763 struct inode
*btree_inode
;
764 btree_inode
= root
->fs_info
->btree_inode
;
765 return filemap_write_and_wait(btree_inode
->i_mapping
);
767 return btrfs_write_and_wait_marked_extents(root
,
768 &trans
->transaction
->dirty_pages
,
773 * this is used to update the root pointer in the tree of tree roots.
775 * But, in the case of the extent allocation tree, updating the root
776 * pointer may allocate blocks which may change the root of the extent
779 * So, this loops and repeats and makes sure the cowonly root didn't
780 * change while the root pointer was being updated in the metadata.
782 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
783 struct btrfs_root
*root
)
788 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
790 old_root_used
= btrfs_root_used(&root
->root_item
);
791 btrfs_write_dirty_block_groups(trans
, root
);
794 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
795 if (old_root_bytenr
== root
->node
->start
&&
796 old_root_used
== btrfs_root_used(&root
->root_item
))
799 btrfs_set_root_node(&root
->root_item
, root
->node
);
800 ret
= btrfs_update_root(trans
, tree_root
,
806 old_root_used
= btrfs_root_used(&root
->root_item
);
807 ret
= btrfs_write_dirty_block_groups(trans
, root
);
812 if (root
!= root
->fs_info
->extent_root
)
813 switch_commit_root(root
);
819 * update all the cowonly tree roots on disk
821 * The error handling in this function may not be obvious. Any of the
822 * failures will cause the file system to go offline. We still need
823 * to clean up the delayed refs.
825 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
,
826 struct btrfs_root
*root
)
828 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
829 struct list_head
*next
;
830 struct extent_buffer
*eb
;
833 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
837 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
838 ret
= btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
,
840 btrfs_tree_unlock(eb
);
841 free_extent_buffer(eb
);
846 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
850 ret
= btrfs_run_dev_stats(trans
, root
->fs_info
);
853 ret
= btrfs_run_qgroups(trans
, root
->fs_info
);
856 /* run_qgroups might have added some more refs */
857 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
860 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
861 next
= fs_info
->dirty_cowonly_roots
.next
;
863 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
865 ret
= update_cowonly_root(trans
, root
);
870 down_write(&fs_info
->extent_commit_sem
);
871 switch_commit_root(fs_info
->extent_root
);
872 up_write(&fs_info
->extent_commit_sem
);
878 * dead roots are old snapshots that need to be deleted. This allocates
879 * a dirty root struct and adds it into the list of dead roots that need to
882 int btrfs_add_dead_root(struct btrfs_root
*root
)
884 spin_lock(&root
->fs_info
->trans_lock
);
885 list_add(&root
->root_list
, &root
->fs_info
->dead_roots
);
886 spin_unlock(&root
->fs_info
->trans_lock
);
891 * update all the cowonly tree roots on disk
893 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
,
894 struct btrfs_root
*root
)
896 struct btrfs_root
*gang
[8];
897 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
902 spin_lock(&fs_info
->fs_roots_radix_lock
);
904 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
907 BTRFS_ROOT_TRANS_TAG
);
910 for (i
= 0; i
< ret
; i
++) {
912 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
913 (unsigned long)root
->root_key
.objectid
,
914 BTRFS_ROOT_TRANS_TAG
);
915 spin_unlock(&fs_info
->fs_roots_radix_lock
);
917 btrfs_free_log(trans
, root
);
918 btrfs_update_reloc_root(trans
, root
);
919 btrfs_orphan_commit_root(trans
, root
);
921 btrfs_save_ino_cache(root
, trans
);
923 /* see comments in should_cow_block() */
927 if (root
->commit_root
!= root
->node
) {
928 mutex_lock(&root
->fs_commit_mutex
);
929 switch_commit_root(root
);
930 btrfs_unpin_free_ino(root
);
931 mutex_unlock(&root
->fs_commit_mutex
);
933 btrfs_set_root_node(&root
->root_item
,
937 err
= btrfs_update_root(trans
, fs_info
->tree_root
,
940 spin_lock(&fs_info
->fs_roots_radix_lock
);
945 spin_unlock(&fs_info
->fs_roots_radix_lock
);
950 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
951 * otherwise every leaf in the btree is read and defragged.
953 int btrfs_defrag_root(struct btrfs_root
*root
, int cacheonly
)
955 struct btrfs_fs_info
*info
= root
->fs_info
;
956 struct btrfs_trans_handle
*trans
;
960 if (xchg(&root
->defrag_running
, 1))
964 trans
= btrfs_start_transaction(root
, 0);
966 return PTR_ERR(trans
);
968 ret
= btrfs_defrag_leaves(trans
, root
, cacheonly
);
970 nr
= trans
->blocks_used
;
971 btrfs_end_transaction(trans
, root
);
972 btrfs_btree_balance_dirty(info
->tree_root
, nr
);
975 if (btrfs_fs_closing(root
->fs_info
) || ret
!= -EAGAIN
)
978 root
->defrag_running
= 0;
983 * new snapshots need to be created at a very specific time in the
984 * transaction commit. This does the actual creation
986 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
987 struct btrfs_fs_info
*fs_info
,
988 struct btrfs_pending_snapshot
*pending
)
990 struct btrfs_key key
;
991 struct btrfs_root_item
*new_root_item
;
992 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
993 struct btrfs_root
*root
= pending
->root
;
994 struct btrfs_root
*parent_root
;
995 struct btrfs_block_rsv
*rsv
;
996 struct inode
*parent_inode
;
997 struct btrfs_path
*path
;
998 struct btrfs_dir_item
*dir_item
;
999 struct dentry
*parent
;
1000 struct dentry
*dentry
;
1001 struct extent_buffer
*tmp
;
1002 struct extent_buffer
*old
;
1003 struct timespec cur_time
= CURRENT_TIME
;
1011 path
= btrfs_alloc_path();
1013 ret
= pending
->error
= -ENOMEM
;
1014 goto path_alloc_fail
;
1017 new_root_item
= kmalloc(sizeof(*new_root_item
), GFP_NOFS
);
1018 if (!new_root_item
) {
1019 ret
= pending
->error
= -ENOMEM
;
1020 goto root_item_alloc_fail
;
1023 ret
= btrfs_find_free_objectid(tree_root
, &objectid
);
1025 pending
->error
= ret
;
1026 goto no_free_objectid
;
1029 btrfs_reloc_pre_snapshot(trans
, pending
, &to_reserve
);
1031 if (to_reserve
> 0) {
1032 ret
= btrfs_block_rsv_add(root
, &pending
->block_rsv
,
1034 BTRFS_RESERVE_NO_FLUSH
);
1036 pending
->error
= ret
;
1037 goto no_free_objectid
;
1041 ret
= btrfs_qgroup_inherit(trans
, fs_info
, root
->root_key
.objectid
,
1042 objectid
, pending
->inherit
);
1044 pending
->error
= ret
;
1045 goto no_free_objectid
;
1048 key
.objectid
= objectid
;
1049 key
.offset
= (u64
)-1;
1050 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1052 rsv
= trans
->block_rsv
;
1053 trans
->block_rsv
= &pending
->block_rsv
;
1055 dentry
= pending
->dentry
;
1056 parent
= dget_parent(dentry
);
1057 parent_inode
= parent
->d_inode
;
1058 parent_root
= BTRFS_I(parent_inode
)->root
;
1059 record_root_in_trans(trans
, parent_root
);
1062 * insert the directory item
1064 ret
= btrfs_set_inode_index(parent_inode
, &index
);
1065 BUG_ON(ret
); /* -ENOMEM */
1067 /* check if there is a file/dir which has the same name. */
1068 dir_item
= btrfs_lookup_dir_item(NULL
, parent_root
, path
,
1069 btrfs_ino(parent_inode
),
1070 dentry
->d_name
.name
,
1071 dentry
->d_name
.len
, 0);
1072 if (dir_item
!= NULL
&& !IS_ERR(dir_item
)) {
1073 pending
->error
= -EEXIST
;
1075 } else if (IS_ERR(dir_item
)) {
1076 ret
= PTR_ERR(dir_item
);
1077 btrfs_abort_transaction(trans
, root
, ret
);
1080 btrfs_release_path(path
);
1083 * pull in the delayed directory update
1084 * and the delayed inode item
1085 * otherwise we corrupt the FS during
1088 ret
= btrfs_run_delayed_items(trans
, root
);
1089 if (ret
) { /* Transaction aborted */
1090 btrfs_abort_transaction(trans
, root
, ret
);
1094 record_root_in_trans(trans
, root
);
1095 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
1096 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
1097 btrfs_check_and_init_root_item(new_root_item
);
1099 root_flags
= btrfs_root_flags(new_root_item
);
1100 if (pending
->readonly
)
1101 root_flags
|= BTRFS_ROOT_SUBVOL_RDONLY
;
1103 root_flags
&= ~BTRFS_ROOT_SUBVOL_RDONLY
;
1104 btrfs_set_root_flags(new_root_item
, root_flags
);
1106 btrfs_set_root_generation_v2(new_root_item
,
1108 uuid_le_gen(&new_uuid
);
1109 memcpy(new_root_item
->uuid
, new_uuid
.b
, BTRFS_UUID_SIZE
);
1110 memcpy(new_root_item
->parent_uuid
, root
->root_item
.uuid
,
1112 new_root_item
->otime
.sec
= cpu_to_le64(cur_time
.tv_sec
);
1113 new_root_item
->otime
.nsec
= cpu_to_le32(cur_time
.tv_nsec
);
1114 btrfs_set_root_otransid(new_root_item
, trans
->transid
);
1115 memset(&new_root_item
->stime
, 0, sizeof(new_root_item
->stime
));
1116 memset(&new_root_item
->rtime
, 0, sizeof(new_root_item
->rtime
));
1117 btrfs_set_root_stransid(new_root_item
, 0);
1118 btrfs_set_root_rtransid(new_root_item
, 0);
1120 old
= btrfs_lock_root_node(root
);
1121 ret
= btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
);
1123 btrfs_tree_unlock(old
);
1124 free_extent_buffer(old
);
1125 btrfs_abort_transaction(trans
, root
, ret
);
1129 btrfs_set_lock_blocking(old
);
1131 ret
= btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
1132 /* clean up in any case */
1133 btrfs_tree_unlock(old
);
1134 free_extent_buffer(old
);
1136 btrfs_abort_transaction(trans
, root
, ret
);
1140 /* see comments in should_cow_block() */
1141 root
->force_cow
= 1;
1144 btrfs_set_root_node(new_root_item
, tmp
);
1145 /* record when the snapshot was created in key.offset */
1146 key
.offset
= trans
->transid
;
1147 ret
= btrfs_insert_root(trans
, tree_root
, &key
, new_root_item
);
1148 btrfs_tree_unlock(tmp
);
1149 free_extent_buffer(tmp
);
1151 btrfs_abort_transaction(trans
, root
, ret
);
1156 * insert root back/forward references
1158 ret
= btrfs_add_root_ref(trans
, tree_root
, objectid
,
1159 parent_root
->root_key
.objectid
,
1160 btrfs_ino(parent_inode
), index
,
1161 dentry
->d_name
.name
, dentry
->d_name
.len
);
1163 btrfs_abort_transaction(trans
, root
, ret
);
1167 key
.offset
= (u64
)-1;
1168 pending
->snap
= btrfs_read_fs_root_no_name(root
->fs_info
, &key
);
1169 if (IS_ERR(pending
->snap
)) {
1170 ret
= PTR_ERR(pending
->snap
);
1171 btrfs_abort_transaction(trans
, root
, ret
);
1175 ret
= btrfs_reloc_post_snapshot(trans
, pending
);
1177 btrfs_abort_transaction(trans
, root
, ret
);
1181 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1183 btrfs_abort_transaction(trans
, root
, ret
);
1187 ret
= btrfs_insert_dir_item(trans
, parent_root
,
1188 dentry
->d_name
.name
, dentry
->d_name
.len
,
1190 BTRFS_FT_DIR
, index
);
1191 /* We have check then name at the beginning, so it is impossible. */
1192 BUG_ON(ret
== -EEXIST
);
1194 btrfs_abort_transaction(trans
, root
, ret
);
1198 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
1199 dentry
->d_name
.len
* 2);
1200 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
1201 ret
= btrfs_update_inode_fallback(trans
, parent_root
, parent_inode
);
1203 btrfs_abort_transaction(trans
, root
, ret
);
1206 trans
->block_rsv
= rsv
;
1208 kfree(new_root_item
);
1209 root_item_alloc_fail
:
1210 btrfs_free_path(path
);
1212 btrfs_block_rsv_release(root
, &pending
->block_rsv
, (u64
)-1);
1217 * create all the snapshots we've scheduled for creation
1219 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
,
1220 struct btrfs_fs_info
*fs_info
)
1222 struct btrfs_pending_snapshot
*pending
;
1223 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
1225 list_for_each_entry(pending
, head
, list
)
1226 create_pending_snapshot(trans
, fs_info
, pending
);
1230 static void update_super_roots(struct btrfs_root
*root
)
1232 struct btrfs_root_item
*root_item
;
1233 struct btrfs_super_block
*super
;
1235 super
= root
->fs_info
->super_copy
;
1237 root_item
= &root
->fs_info
->chunk_root
->root_item
;
1238 super
->chunk_root
= root_item
->bytenr
;
1239 super
->chunk_root_generation
= root_item
->generation
;
1240 super
->chunk_root_level
= root_item
->level
;
1242 root_item
= &root
->fs_info
->tree_root
->root_item
;
1243 super
->root
= root_item
->bytenr
;
1244 super
->generation
= root_item
->generation
;
1245 super
->root_level
= root_item
->level
;
1246 if (btrfs_test_opt(root
, SPACE_CACHE
))
1247 super
->cache_generation
= root_item
->generation
;
1250 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
1253 spin_lock(&info
->trans_lock
);
1254 if (info
->running_transaction
)
1255 ret
= info
->running_transaction
->in_commit
;
1256 spin_unlock(&info
->trans_lock
);
1260 int btrfs_transaction_blocked(struct btrfs_fs_info
*info
)
1263 spin_lock(&info
->trans_lock
);
1264 if (info
->running_transaction
)
1265 ret
= info
->running_transaction
->blocked
;
1266 spin_unlock(&info
->trans_lock
);
1271 * wait for the current transaction commit to start and block subsequent
1274 static void wait_current_trans_commit_start(struct btrfs_root
*root
,
1275 struct btrfs_transaction
*trans
)
1277 wait_event(root
->fs_info
->transaction_blocked_wait
, trans
->in_commit
);
1281 * wait for the current transaction to start and then become unblocked.
1284 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root
*root
,
1285 struct btrfs_transaction
*trans
)
1287 wait_event(root
->fs_info
->transaction_wait
,
1288 trans
->commit_done
|| (trans
->in_commit
&& !trans
->blocked
));
1292 * commit transactions asynchronously. once btrfs_commit_transaction_async
1293 * returns, any subsequent transaction will not be allowed to join.
1295 struct btrfs_async_commit
{
1296 struct btrfs_trans_handle
*newtrans
;
1297 struct btrfs_root
*root
;
1298 struct delayed_work work
;
1301 static void do_async_commit(struct work_struct
*work
)
1303 struct btrfs_async_commit
*ac
=
1304 container_of(work
, struct btrfs_async_commit
, work
.work
);
1307 * We've got freeze protection passed with the transaction.
1308 * Tell lockdep about it.
1311 &ac
->root
->fs_info
->sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
1314 current
->journal_info
= ac
->newtrans
;
1316 btrfs_commit_transaction(ac
->newtrans
, ac
->root
);
1320 int btrfs_commit_transaction_async(struct btrfs_trans_handle
*trans
,
1321 struct btrfs_root
*root
,
1322 int wait_for_unblock
)
1324 struct btrfs_async_commit
*ac
;
1325 struct btrfs_transaction
*cur_trans
;
1327 ac
= kmalloc(sizeof(*ac
), GFP_NOFS
);
1331 INIT_DELAYED_WORK(&ac
->work
, do_async_commit
);
1333 ac
->newtrans
= btrfs_join_transaction(root
);
1334 if (IS_ERR(ac
->newtrans
)) {
1335 int err
= PTR_ERR(ac
->newtrans
);
1340 /* take transaction reference */
1341 cur_trans
= trans
->transaction
;
1342 atomic_inc(&cur_trans
->use_count
);
1344 btrfs_end_transaction(trans
, root
);
1347 * Tell lockdep we've released the freeze rwsem, since the
1348 * async commit thread will be the one to unlock it.
1350 rwsem_release(&root
->fs_info
->sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
1353 schedule_delayed_work(&ac
->work
, 0);
1355 /* wait for transaction to start and unblock */
1356 if (wait_for_unblock
)
1357 wait_current_trans_commit_start_and_unblock(root
, cur_trans
);
1359 wait_current_trans_commit_start(root
, cur_trans
);
1361 if (current
->journal_info
== trans
)
1362 current
->journal_info
= NULL
;
1364 put_transaction(cur_trans
);
1369 static void cleanup_transaction(struct btrfs_trans_handle
*trans
,
1370 struct btrfs_root
*root
, int err
)
1372 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1374 WARN_ON(trans
->use_count
> 1);
1376 btrfs_abort_transaction(trans
, root
, err
);
1378 spin_lock(&root
->fs_info
->trans_lock
);
1379 list_del_init(&cur_trans
->list
);
1380 if (cur_trans
== root
->fs_info
->running_transaction
) {
1381 root
->fs_info
->running_transaction
= NULL
;
1382 root
->fs_info
->trans_no_join
= 0;
1384 spin_unlock(&root
->fs_info
->trans_lock
);
1386 btrfs_cleanup_one_transaction(trans
->transaction
, root
);
1388 put_transaction(cur_trans
);
1389 put_transaction(cur_trans
);
1391 trace_btrfs_transaction_commit(root
);
1393 btrfs_scrub_continue(root
);
1395 if (current
->journal_info
== trans
)
1396 current
->journal_info
= NULL
;
1398 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1402 * btrfs_transaction state sequence:
1403 * in_commit = 0, blocked = 0 (initial)
1404 * in_commit = 1, blocked = 1
1408 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
,
1409 struct btrfs_root
*root
)
1411 unsigned long joined
= 0;
1412 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1413 struct btrfs_transaction
*prev_trans
= NULL
;
1416 int should_grow
= 0;
1417 unsigned long now
= get_seconds();
1418 int flush_on_commit
= btrfs_test_opt(root
, FLUSHONCOMMIT
);
1420 btrfs_run_ordered_operations(root
, 0);
1422 if (cur_trans
->aborted
)
1423 goto cleanup_transaction
;
1425 /* make a pass through all the delayed refs we have so far
1426 * any runnings procs may add more while we are here
1428 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1430 goto cleanup_transaction
;
1432 btrfs_trans_release_metadata(trans
, root
);
1433 trans
->block_rsv
= NULL
;
1435 cur_trans
= trans
->transaction
;
1438 * set the flushing flag so procs in this transaction have to
1439 * start sending their work down.
1441 cur_trans
->delayed_refs
.flushing
= 1;
1443 if (!list_empty(&trans
->new_bgs
))
1444 btrfs_create_pending_block_groups(trans
, root
);
1446 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1448 goto cleanup_transaction
;
1450 spin_lock(&cur_trans
->commit_lock
);
1451 if (cur_trans
->in_commit
) {
1452 spin_unlock(&cur_trans
->commit_lock
);
1453 atomic_inc(&cur_trans
->use_count
);
1454 ret
= btrfs_end_transaction(trans
, root
);
1456 wait_for_commit(root
, cur_trans
);
1458 put_transaction(cur_trans
);
1463 trans
->transaction
->in_commit
= 1;
1464 trans
->transaction
->blocked
= 1;
1465 spin_unlock(&cur_trans
->commit_lock
);
1466 wake_up(&root
->fs_info
->transaction_blocked_wait
);
1468 spin_lock(&root
->fs_info
->trans_lock
);
1469 if (cur_trans
->list
.prev
!= &root
->fs_info
->trans_list
) {
1470 prev_trans
= list_entry(cur_trans
->list
.prev
,
1471 struct btrfs_transaction
, list
);
1472 if (!prev_trans
->commit_done
) {
1473 atomic_inc(&prev_trans
->use_count
);
1474 spin_unlock(&root
->fs_info
->trans_lock
);
1476 wait_for_commit(root
, prev_trans
);
1478 put_transaction(prev_trans
);
1480 spin_unlock(&root
->fs_info
->trans_lock
);
1483 spin_unlock(&root
->fs_info
->trans_lock
);
1486 if (!btrfs_test_opt(root
, SSD
) &&
1487 (now
< cur_trans
->start_time
|| now
- cur_trans
->start_time
< 1))
1491 int snap_pending
= 0;
1493 joined
= cur_trans
->num_joined
;
1494 if (!list_empty(&trans
->transaction
->pending_snapshots
))
1497 WARN_ON(cur_trans
!= trans
->transaction
);
1499 if (flush_on_commit
|| snap_pending
) {
1500 ret
= btrfs_start_delalloc_inodes(root
, 1);
1502 btrfs_abort_transaction(trans
, root
, ret
);
1503 goto cleanup_transaction
;
1505 btrfs_wait_ordered_extents(root
, 1);
1508 ret
= btrfs_run_delayed_items(trans
, root
);
1510 goto cleanup_transaction
;
1513 * running the delayed items may have added new refs. account
1514 * them now so that they hinder processing of more delayed refs
1515 * as little as possible.
1517 btrfs_delayed_refs_qgroup_accounting(trans
, root
->fs_info
);
1520 * rename don't use btrfs_join_transaction, so, once we
1521 * set the transaction to blocked above, we aren't going
1522 * to get any new ordered operations. We can safely run
1523 * it here and no for sure that nothing new will be added
1526 btrfs_run_ordered_operations(root
, 1);
1528 prepare_to_wait(&cur_trans
->writer_wait
, &wait
,
1529 TASK_UNINTERRUPTIBLE
);
1531 if (atomic_read(&cur_trans
->num_writers
) > 1)
1532 schedule_timeout(MAX_SCHEDULE_TIMEOUT
);
1533 else if (should_grow
)
1534 schedule_timeout(1);
1536 finish_wait(&cur_trans
->writer_wait
, &wait
);
1537 } while (atomic_read(&cur_trans
->num_writers
) > 1 ||
1538 (should_grow
&& cur_trans
->num_joined
!= joined
));
1541 * Ok now we need to make sure to block out any other joins while we
1542 * commit the transaction. We could have started a join before setting
1543 * no_join so make sure to wait for num_writers to == 1 again.
1545 spin_lock(&root
->fs_info
->trans_lock
);
1546 root
->fs_info
->trans_no_join
= 1;
1547 spin_unlock(&root
->fs_info
->trans_lock
);
1548 wait_event(cur_trans
->writer_wait
,
1549 atomic_read(&cur_trans
->num_writers
) == 1);
1552 * the reloc mutex makes sure that we stop
1553 * the balancing code from coming in and moving
1554 * extents around in the middle of the commit
1556 mutex_lock(&root
->fs_info
->reloc_mutex
);
1559 * We needn't worry about the delayed items because we will
1560 * deal with them in create_pending_snapshot(), which is the
1561 * core function of the snapshot creation.
1563 ret
= create_pending_snapshots(trans
, root
->fs_info
);
1565 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1566 goto cleanup_transaction
;
1570 * We insert the dir indexes of the snapshots and update the inode
1571 * of the snapshots' parents after the snapshot creation, so there
1572 * are some delayed items which are not dealt with. Now deal with
1575 * We needn't worry that this operation will corrupt the snapshots,
1576 * because all the tree which are snapshoted will be forced to COW
1577 * the nodes and leaves.
1579 ret
= btrfs_run_delayed_items(trans
, root
);
1581 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1582 goto cleanup_transaction
;
1585 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1587 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1588 goto cleanup_transaction
;
1592 * make sure none of the code above managed to slip in a
1595 btrfs_assert_delayed_root_empty(root
);
1597 WARN_ON(cur_trans
!= trans
->transaction
);
1599 btrfs_scrub_pause(root
);
1600 /* btrfs_commit_tree_roots is responsible for getting the
1601 * various roots consistent with each other. Every pointer
1602 * in the tree of tree roots has to point to the most up to date
1603 * root for every subvolume and other tree. So, we have to keep
1604 * the tree logging code from jumping in and changing any
1607 * At this point in the commit, there can't be any tree-log
1608 * writers, but a little lower down we drop the trans mutex
1609 * and let new people in. By holding the tree_log_mutex
1610 * from now until after the super is written, we avoid races
1611 * with the tree-log code.
1613 mutex_lock(&root
->fs_info
->tree_log_mutex
);
1615 ret
= commit_fs_roots(trans
, root
);
1617 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1618 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1619 goto cleanup_transaction
;
1622 /* commit_fs_roots gets rid of all the tree log roots, it is now
1623 * safe to free the root of tree log roots
1625 btrfs_free_log_root_tree(trans
, root
->fs_info
);
1627 ret
= commit_cowonly_roots(trans
, root
);
1629 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1630 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1631 goto cleanup_transaction
;
1634 btrfs_prepare_extent_commit(trans
, root
);
1636 cur_trans
= root
->fs_info
->running_transaction
;
1638 btrfs_set_root_node(&root
->fs_info
->tree_root
->root_item
,
1639 root
->fs_info
->tree_root
->node
);
1640 switch_commit_root(root
->fs_info
->tree_root
);
1642 btrfs_set_root_node(&root
->fs_info
->chunk_root
->root_item
,
1643 root
->fs_info
->chunk_root
->node
);
1644 switch_commit_root(root
->fs_info
->chunk_root
);
1646 assert_qgroups_uptodate(trans
);
1647 update_super_roots(root
);
1649 if (!root
->fs_info
->log_root_recovering
) {
1650 btrfs_set_super_log_root(root
->fs_info
->super_copy
, 0);
1651 btrfs_set_super_log_root_level(root
->fs_info
->super_copy
, 0);
1654 memcpy(root
->fs_info
->super_for_commit
, root
->fs_info
->super_copy
,
1655 sizeof(*root
->fs_info
->super_copy
));
1657 trans
->transaction
->blocked
= 0;
1658 spin_lock(&root
->fs_info
->trans_lock
);
1659 root
->fs_info
->running_transaction
= NULL
;
1660 root
->fs_info
->trans_no_join
= 0;
1661 spin_unlock(&root
->fs_info
->trans_lock
);
1662 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1664 wake_up(&root
->fs_info
->transaction_wait
);
1666 ret
= btrfs_write_and_wait_transaction(trans
, root
);
1668 btrfs_error(root
->fs_info
, ret
,
1669 "Error while writing out transaction.");
1670 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1671 goto cleanup_transaction
;
1674 ret
= write_ctree_super(trans
, root
, 0);
1676 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1677 goto cleanup_transaction
;
1681 * the super is written, we can safely allow the tree-loggers
1682 * to go about their business
1684 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1686 btrfs_finish_extent_commit(trans
, root
);
1688 cur_trans
->commit_done
= 1;
1690 root
->fs_info
->last_trans_committed
= cur_trans
->transid
;
1692 wake_up(&cur_trans
->commit_wait
);
1694 spin_lock(&root
->fs_info
->trans_lock
);
1695 list_del_init(&cur_trans
->list
);
1696 spin_unlock(&root
->fs_info
->trans_lock
);
1698 put_transaction(cur_trans
);
1699 put_transaction(cur_trans
);
1701 if (trans
->type
< TRANS_JOIN_NOLOCK
)
1702 sb_end_intwrite(root
->fs_info
->sb
);
1704 trace_btrfs_transaction_commit(root
);
1706 btrfs_scrub_continue(root
);
1708 if (current
->journal_info
== trans
)
1709 current
->journal_info
= NULL
;
1711 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1713 if (current
!= root
->fs_info
->transaction_kthread
)
1714 btrfs_run_delayed_iputs(root
);
1718 cleanup_transaction
:
1719 btrfs_trans_release_metadata(trans
, root
);
1720 trans
->block_rsv
= NULL
;
1721 btrfs_printk(root
->fs_info
, "Skipping commit of aborted transaction.\n");
1723 if (current
->journal_info
== trans
)
1724 current
->journal_info
= NULL
;
1725 cleanup_transaction(trans
, root
, ret
);
1731 * interface function to delete all the snapshots we have scheduled for deletion
1733 int btrfs_clean_old_snapshots(struct btrfs_root
*root
)
1736 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1738 spin_lock(&fs_info
->trans_lock
);
1739 list_splice_init(&fs_info
->dead_roots
, &list
);
1740 spin_unlock(&fs_info
->trans_lock
);
1742 while (!list_empty(&list
)) {
1745 root
= list_entry(list
.next
, struct btrfs_root
, root_list
);
1746 list_del(&root
->root_list
);
1748 btrfs_kill_all_delayed_nodes(root
);
1750 if (btrfs_header_backref_rev(root
->node
) <
1751 BTRFS_MIXED_BACKREF_REV
)
1752 ret
= btrfs_drop_snapshot(root
, NULL
, 0, 0);
1754 ret
=btrfs_drop_snapshot(root
, NULL
, 1, 0);