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"
33 #include "dev-replace.h"
35 #define BTRFS_ROOT_TRANS_TAG 0
37 void put_transaction(struct btrfs_transaction
*transaction
)
39 WARN_ON(atomic_read(&transaction
->use_count
) == 0);
40 if (atomic_dec_and_test(&transaction
->use_count
)) {
41 BUG_ON(!list_empty(&transaction
->list
));
42 WARN_ON(transaction
->delayed_refs
.root
.rb_node
);
43 memset(transaction
, 0, sizeof(*transaction
));
44 kmem_cache_free(btrfs_transaction_cachep
, transaction
);
48 static noinline
void switch_commit_root(struct btrfs_root
*root
)
50 free_extent_buffer(root
->commit_root
);
51 root
->commit_root
= btrfs_root_node(root
);
55 * either allocate a new transaction or hop into the existing one
57 static noinline
int join_transaction(struct btrfs_root
*root
, int type
)
59 struct btrfs_transaction
*cur_trans
;
60 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
62 spin_lock(&fs_info
->trans_lock
);
64 /* The file system has been taken offline. No new transactions. */
65 if (fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
) {
66 spin_unlock(&fs_info
->trans_lock
);
70 if (fs_info
->trans_no_join
) {
72 * If we are JOIN_NOLOCK we're already committing a current
73 * transaction, we just need a handle to deal with something
74 * when committing the transaction, such as inode cache and
75 * space cache. It is a special case.
77 if (type
!= TRANS_JOIN_NOLOCK
) {
78 spin_unlock(&fs_info
->trans_lock
);
83 cur_trans
= fs_info
->running_transaction
;
85 if (cur_trans
->aborted
) {
86 spin_unlock(&fs_info
->trans_lock
);
87 return cur_trans
->aborted
;
89 atomic_inc(&cur_trans
->use_count
);
90 atomic_inc(&cur_trans
->num_writers
);
91 cur_trans
->num_joined
++;
92 spin_unlock(&fs_info
->trans_lock
);
95 spin_unlock(&fs_info
->trans_lock
);
98 * If we are ATTACH, we just want to catch the current transaction,
99 * and commit it. If there is no transaction, just return ENOENT.
101 if (type
== TRANS_ATTACH
)
104 cur_trans
= kmem_cache_alloc(btrfs_transaction_cachep
, GFP_NOFS
);
108 spin_lock(&fs_info
->trans_lock
);
109 if (fs_info
->running_transaction
) {
111 * someone started a transaction after we unlocked. Make sure
112 * to redo the trans_no_join checks above
114 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
115 cur_trans
= fs_info
->running_transaction
;
117 } else if (fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
) {
118 spin_unlock(&fs_info
->trans_lock
);
119 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
123 atomic_set(&cur_trans
->num_writers
, 1);
124 cur_trans
->num_joined
= 0;
125 init_waitqueue_head(&cur_trans
->writer_wait
);
126 init_waitqueue_head(&cur_trans
->commit_wait
);
127 cur_trans
->in_commit
= 0;
128 cur_trans
->blocked
= 0;
130 * One for this trans handle, one so it will live on until we
131 * commit the transaction.
133 atomic_set(&cur_trans
->use_count
, 2);
134 cur_trans
->commit_done
= 0;
135 cur_trans
->start_time
= get_seconds();
137 cur_trans
->delayed_refs
.root
= RB_ROOT
;
138 cur_trans
->delayed_refs
.num_entries
= 0;
139 cur_trans
->delayed_refs
.num_heads_ready
= 0;
140 cur_trans
->delayed_refs
.num_heads
= 0;
141 cur_trans
->delayed_refs
.flushing
= 0;
142 cur_trans
->delayed_refs
.run_delayed_start
= 0;
145 * although the tree mod log is per file system and not per transaction,
146 * the log must never go across transaction boundaries.
149 if (!list_empty(&fs_info
->tree_mod_seq_list
))
150 WARN(1, KERN_ERR
"btrfs: tree_mod_seq_list not empty when "
151 "creating a fresh transaction\n");
152 if (!RB_EMPTY_ROOT(&fs_info
->tree_mod_log
))
153 WARN(1, KERN_ERR
"btrfs: tree_mod_log rb tree not empty when "
154 "creating a fresh transaction\n");
155 atomic_set(&fs_info
->tree_mod_seq
, 0);
157 spin_lock_init(&cur_trans
->commit_lock
);
158 spin_lock_init(&cur_trans
->delayed_refs
.lock
);
160 INIT_LIST_HEAD(&cur_trans
->pending_snapshots
);
161 list_add_tail(&cur_trans
->list
, &fs_info
->trans_list
);
162 extent_io_tree_init(&cur_trans
->dirty_pages
,
163 fs_info
->btree_inode
->i_mapping
);
164 fs_info
->generation
++;
165 cur_trans
->transid
= fs_info
->generation
;
166 fs_info
->running_transaction
= cur_trans
;
167 cur_trans
->aborted
= 0;
168 spin_unlock(&fs_info
->trans_lock
);
174 * this does all the record keeping required to make sure that a reference
175 * counted root is properly recorded in a given transaction. This is required
176 * to make sure the old root from before we joined the transaction is deleted
177 * when the transaction commits
179 static int record_root_in_trans(struct btrfs_trans_handle
*trans
,
180 struct btrfs_root
*root
)
182 if (root
->ref_cows
&& root
->last_trans
< trans
->transid
) {
183 WARN_ON(root
== root
->fs_info
->extent_root
);
184 WARN_ON(root
->commit_root
!= root
->node
);
187 * see below for in_trans_setup usage rules
188 * we have the reloc mutex held now, so there
189 * is only one writer in this function
191 root
->in_trans_setup
= 1;
193 /* make sure readers find in_trans_setup before
194 * they find our root->last_trans update
198 spin_lock(&root
->fs_info
->fs_roots_radix_lock
);
199 if (root
->last_trans
== trans
->transid
) {
200 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
203 radix_tree_tag_set(&root
->fs_info
->fs_roots_radix
,
204 (unsigned long)root
->root_key
.objectid
,
205 BTRFS_ROOT_TRANS_TAG
);
206 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
207 root
->last_trans
= trans
->transid
;
209 /* this is pretty tricky. We don't want to
210 * take the relocation lock in btrfs_record_root_in_trans
211 * unless we're really doing the first setup for this root in
214 * Normally we'd use root->last_trans as a flag to decide
215 * if we want to take the expensive mutex.
217 * But, we have to set root->last_trans before we
218 * init the relocation root, otherwise, we trip over warnings
219 * in ctree.c. The solution used here is to flag ourselves
220 * with root->in_trans_setup. When this is 1, we're still
221 * fixing up the reloc trees and everyone must wait.
223 * When this is zero, they can trust root->last_trans and fly
224 * through btrfs_record_root_in_trans without having to take the
225 * lock. smp_wmb() makes sure that all the writes above are
226 * done before we pop in the zero below
228 btrfs_init_reloc_root(trans
, root
);
230 root
->in_trans_setup
= 0;
236 int btrfs_record_root_in_trans(struct btrfs_trans_handle
*trans
,
237 struct btrfs_root
*root
)
243 * see record_root_in_trans for comments about in_trans_setup usage
247 if (root
->last_trans
== trans
->transid
&&
248 !root
->in_trans_setup
)
251 mutex_lock(&root
->fs_info
->reloc_mutex
);
252 record_root_in_trans(trans
, root
);
253 mutex_unlock(&root
->fs_info
->reloc_mutex
);
258 /* wait for commit against the current transaction to become unblocked
259 * when this is done, it is safe to start a new transaction, but the current
260 * transaction might not be fully on disk.
262 static void wait_current_trans(struct btrfs_root
*root
)
264 struct btrfs_transaction
*cur_trans
;
266 spin_lock(&root
->fs_info
->trans_lock
);
267 cur_trans
= root
->fs_info
->running_transaction
;
268 if (cur_trans
&& cur_trans
->blocked
) {
269 atomic_inc(&cur_trans
->use_count
);
270 spin_unlock(&root
->fs_info
->trans_lock
);
272 wait_event(root
->fs_info
->transaction_wait
,
273 !cur_trans
->blocked
);
274 put_transaction(cur_trans
);
276 spin_unlock(&root
->fs_info
->trans_lock
);
280 static int may_wait_transaction(struct btrfs_root
*root
, int type
)
282 if (root
->fs_info
->log_root_recovering
)
285 if (type
== TRANS_USERSPACE
)
288 if (type
== TRANS_START
&&
289 !atomic_read(&root
->fs_info
->open_ioctl_trans
))
295 static struct btrfs_trans_handle
*
296 start_transaction(struct btrfs_root
*root
, u64 num_items
, int type
,
297 enum btrfs_reserve_flush_enum flush
)
299 struct btrfs_trans_handle
*h
;
300 struct btrfs_transaction
*cur_trans
;
303 u64 qgroup_reserved
= 0;
305 if (root
->fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
)
306 return ERR_PTR(-EROFS
);
308 if (current
->journal_info
) {
309 WARN_ON(type
!= TRANS_JOIN
&& type
!= TRANS_JOIN_NOLOCK
);
310 h
= current
->journal_info
;
312 WARN_ON(h
->use_count
> 2);
313 h
->orig_rsv
= h
->block_rsv
;
319 * Do the reservation before we join the transaction so we can do all
320 * the appropriate flushing if need be.
322 if (num_items
> 0 && root
!= root
->fs_info
->chunk_root
) {
323 if (root
->fs_info
->quota_enabled
&&
324 is_fstree(root
->root_key
.objectid
)) {
325 qgroup_reserved
= num_items
* root
->leafsize
;
326 ret
= btrfs_qgroup_reserve(root
, qgroup_reserved
);
331 num_bytes
= btrfs_calc_trans_metadata_size(root
, num_items
);
332 ret
= btrfs_block_rsv_add(root
,
333 &root
->fs_info
->trans_block_rsv
,
339 h
= kmem_cache_alloc(btrfs_trans_handle_cachep
, GFP_NOFS
);
341 return ERR_PTR(-ENOMEM
);
344 * If we are JOIN_NOLOCK we're already committing a transaction and
345 * waiting on this guy, so we don't need to do the sb_start_intwrite
346 * because we're already holding a ref. We need this because we could
347 * have raced in and did an fsync() on a file which can kick a commit
348 * and then we deadlock with somebody doing a freeze.
350 * If we are ATTACH, it means we just want to catch the current
351 * transaction and commit it, so we needn't do sb_start_intwrite().
353 if (type
< TRANS_JOIN_NOLOCK
)
354 sb_start_intwrite(root
->fs_info
->sb
);
356 if (may_wait_transaction(root
, type
))
357 wait_current_trans(root
);
360 ret
= join_transaction(root
, type
);
362 wait_current_trans(root
);
363 } while (ret
== -EBUSY
);
366 /* We must get the transaction if we are JOIN_NOLOCK. */
367 BUG_ON(type
== TRANS_JOIN_NOLOCK
);
369 if (type
< TRANS_JOIN_NOLOCK
)
370 sb_end_intwrite(root
->fs_info
->sb
);
371 kmem_cache_free(btrfs_trans_handle_cachep
, h
);
375 cur_trans
= root
->fs_info
->running_transaction
;
377 h
->transid
= cur_trans
->transid
;
378 h
->transaction
= cur_trans
;
380 h
->bytes_reserved
= 0;
382 h
->delayed_ref_updates
= 0;
388 h
->qgroup_reserved
= qgroup_reserved
;
389 h
->delayed_ref_elem
.seq
= 0;
391 INIT_LIST_HEAD(&h
->qgroup_ref_list
);
392 INIT_LIST_HEAD(&h
->new_bgs
);
395 if (cur_trans
->blocked
&& may_wait_transaction(root
, type
)) {
396 btrfs_commit_transaction(h
, root
);
401 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
402 h
->transid
, num_bytes
, 1);
403 h
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
404 h
->bytes_reserved
= num_bytes
;
408 btrfs_record_root_in_trans(h
, root
);
410 if (!current
->journal_info
&& type
!= TRANS_USERSPACE
)
411 current
->journal_info
= h
;
415 struct btrfs_trans_handle
*btrfs_start_transaction(struct btrfs_root
*root
,
418 return start_transaction(root
, num_items
, TRANS_START
,
419 BTRFS_RESERVE_FLUSH_ALL
);
422 struct btrfs_trans_handle
*btrfs_start_transaction_lflush(
423 struct btrfs_root
*root
, int num_items
)
425 return start_transaction(root
, num_items
, TRANS_START
,
426 BTRFS_RESERVE_FLUSH_LIMIT
);
429 struct btrfs_trans_handle
*btrfs_join_transaction(struct btrfs_root
*root
)
431 return start_transaction(root
, 0, TRANS_JOIN
, 0);
434 struct btrfs_trans_handle
*btrfs_join_transaction_nolock(struct btrfs_root
*root
)
436 return start_transaction(root
, 0, TRANS_JOIN_NOLOCK
, 0);
439 struct btrfs_trans_handle
*btrfs_start_ioctl_transaction(struct btrfs_root
*root
)
441 return start_transaction(root
, 0, TRANS_USERSPACE
, 0);
444 struct btrfs_trans_handle
*btrfs_attach_transaction(struct btrfs_root
*root
)
446 return start_transaction(root
, 0, TRANS_ATTACH
, 0);
449 /* wait for a transaction commit to be fully complete */
450 static noinline
void wait_for_commit(struct btrfs_root
*root
,
451 struct btrfs_transaction
*commit
)
453 wait_event(commit
->commit_wait
, commit
->commit_done
);
456 int btrfs_wait_for_commit(struct btrfs_root
*root
, u64 transid
)
458 struct btrfs_transaction
*cur_trans
= NULL
, *t
;
462 if (transid
<= root
->fs_info
->last_trans_committed
)
466 /* find specified transaction */
467 spin_lock(&root
->fs_info
->trans_lock
);
468 list_for_each_entry(t
, &root
->fs_info
->trans_list
, list
) {
469 if (t
->transid
== transid
) {
471 atomic_inc(&cur_trans
->use_count
);
475 if (t
->transid
> transid
) {
480 spin_unlock(&root
->fs_info
->trans_lock
);
481 /* The specified transaction doesn't exist */
485 /* find newest transaction that is committing | committed */
486 spin_lock(&root
->fs_info
->trans_lock
);
487 list_for_each_entry_reverse(t
, &root
->fs_info
->trans_list
,
493 atomic_inc(&cur_trans
->use_count
);
497 spin_unlock(&root
->fs_info
->trans_lock
);
499 goto out
; /* nothing committing|committed */
502 wait_for_commit(root
, cur_trans
);
503 put_transaction(cur_trans
);
508 void btrfs_throttle(struct btrfs_root
*root
)
510 if (!atomic_read(&root
->fs_info
->open_ioctl_trans
))
511 wait_current_trans(root
);
514 static int should_end_transaction(struct btrfs_trans_handle
*trans
,
515 struct btrfs_root
*root
)
519 ret
= btrfs_block_rsv_check(root
, &root
->fs_info
->global_block_rsv
, 5);
523 int btrfs_should_end_transaction(struct btrfs_trans_handle
*trans
,
524 struct btrfs_root
*root
)
526 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
531 if (cur_trans
->blocked
|| cur_trans
->delayed_refs
.flushing
)
534 updates
= trans
->delayed_ref_updates
;
535 trans
->delayed_ref_updates
= 0;
537 err
= btrfs_run_delayed_refs(trans
, root
, updates
);
538 if (err
) /* Error code will also eval true */
542 return should_end_transaction(trans
, root
);
545 static int __btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
546 struct btrfs_root
*root
, int throttle
)
548 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
549 struct btrfs_fs_info
*info
= root
->fs_info
;
551 int lock
= (trans
->type
!= TRANS_JOIN_NOLOCK
);
554 if (--trans
->use_count
) {
555 trans
->block_rsv
= trans
->orig_rsv
;
560 * do the qgroup accounting as early as possible
562 err
= btrfs_delayed_refs_qgroup_accounting(trans
, info
);
564 btrfs_trans_release_metadata(trans
, root
);
565 trans
->block_rsv
= NULL
;
567 * the same root has to be passed to start_transaction and
568 * end_transaction. Subvolume quota depends on this.
570 WARN_ON(trans
->root
!= root
);
572 if (trans
->qgroup_reserved
) {
573 btrfs_qgroup_free(root
, trans
->qgroup_reserved
);
574 trans
->qgroup_reserved
= 0;
577 if (!list_empty(&trans
->new_bgs
))
578 btrfs_create_pending_block_groups(trans
, root
);
581 unsigned long cur
= trans
->delayed_ref_updates
;
582 trans
->delayed_ref_updates
= 0;
584 trans
->transaction
->delayed_refs
.num_heads_ready
> 64) {
585 trans
->delayed_ref_updates
= 0;
586 btrfs_run_delayed_refs(trans
, root
, cur
);
592 btrfs_trans_release_metadata(trans
, root
);
593 trans
->block_rsv
= NULL
;
595 if (!list_empty(&trans
->new_bgs
))
596 btrfs_create_pending_block_groups(trans
, root
);
598 if (lock
&& !atomic_read(&root
->fs_info
->open_ioctl_trans
) &&
599 should_end_transaction(trans
, root
)) {
600 trans
->transaction
->blocked
= 1;
604 if (lock
&& cur_trans
->blocked
&& !cur_trans
->in_commit
) {
607 * We may race with somebody else here so end up having
608 * to call end_transaction on ourselves again, so inc
612 return btrfs_commit_transaction(trans
, root
);
614 wake_up_process(info
->transaction_kthread
);
618 if (trans
->type
< TRANS_JOIN_NOLOCK
)
619 sb_end_intwrite(root
->fs_info
->sb
);
621 WARN_ON(cur_trans
!= info
->running_transaction
);
622 WARN_ON(atomic_read(&cur_trans
->num_writers
) < 1);
623 atomic_dec(&cur_trans
->num_writers
);
626 if (waitqueue_active(&cur_trans
->writer_wait
))
627 wake_up(&cur_trans
->writer_wait
);
628 put_transaction(cur_trans
);
630 if (current
->journal_info
== trans
)
631 current
->journal_info
= NULL
;
634 btrfs_run_delayed_iputs(root
);
636 if (trans
->aborted
||
637 root
->fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
) {
640 assert_qgroups_uptodate(trans
);
642 memset(trans
, 0, sizeof(*trans
));
643 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
647 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
648 struct btrfs_root
*root
)
652 ret
= __btrfs_end_transaction(trans
, root
, 0);
658 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
,
659 struct btrfs_root
*root
)
663 ret
= __btrfs_end_transaction(trans
, root
, 1);
669 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle
*trans
,
670 struct btrfs_root
*root
)
672 return __btrfs_end_transaction(trans
, root
, 1);
676 * when btree blocks are allocated, they have some corresponding bits set for
677 * them in one of two extent_io trees. This is used to make sure all of
678 * those extents are sent to disk but does not wait on them
680 int btrfs_write_marked_extents(struct btrfs_root
*root
,
681 struct extent_io_tree
*dirty_pages
, int mark
)
685 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
686 struct extent_state
*cached_state
= NULL
;
690 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
691 mark
, &cached_state
)) {
692 convert_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
,
693 mark
, &cached_state
, GFP_NOFS
);
695 err
= filemap_fdatawrite_range(mapping
, start
, end
);
707 * when btree blocks are allocated, they have some corresponding bits set for
708 * them in one of two extent_io trees. This is used to make sure all of
709 * those extents are on disk for transaction or log commit. We wait
710 * on all the pages and clear them from the dirty pages state tree
712 int btrfs_wait_marked_extents(struct btrfs_root
*root
,
713 struct extent_io_tree
*dirty_pages
, int mark
)
717 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
718 struct extent_state
*cached_state
= NULL
;
722 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
723 EXTENT_NEED_WAIT
, &cached_state
)) {
724 clear_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
,
725 0, 0, &cached_state
, GFP_NOFS
);
726 err
= filemap_fdatawait_range(mapping
, start
, end
);
738 * when btree blocks are allocated, they have some corresponding bits set for
739 * them in one of two extent_io trees. This is used to make sure all of
740 * those extents are on disk for transaction or log commit
742 int btrfs_write_and_wait_marked_extents(struct btrfs_root
*root
,
743 struct extent_io_tree
*dirty_pages
, int mark
)
748 ret
= btrfs_write_marked_extents(root
, dirty_pages
, mark
);
749 ret2
= btrfs_wait_marked_extents(root
, dirty_pages
, mark
);
758 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
,
759 struct btrfs_root
*root
)
761 if (!trans
|| !trans
->transaction
) {
762 struct inode
*btree_inode
;
763 btree_inode
= root
->fs_info
->btree_inode
;
764 return filemap_write_and_wait(btree_inode
->i_mapping
);
766 return btrfs_write_and_wait_marked_extents(root
,
767 &trans
->transaction
->dirty_pages
,
772 * this is used to update the root pointer in the tree of tree roots.
774 * But, in the case of the extent allocation tree, updating the root
775 * pointer may allocate blocks which may change the root of the extent
778 * So, this loops and repeats and makes sure the cowonly root didn't
779 * change while the root pointer was being updated in the metadata.
781 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
782 struct btrfs_root
*root
)
787 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
789 old_root_used
= btrfs_root_used(&root
->root_item
);
790 btrfs_write_dirty_block_groups(trans
, root
);
793 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
794 if (old_root_bytenr
== root
->node
->start
&&
795 old_root_used
== btrfs_root_used(&root
->root_item
))
798 btrfs_set_root_node(&root
->root_item
, root
->node
);
799 ret
= btrfs_update_root(trans
, tree_root
,
805 old_root_used
= btrfs_root_used(&root
->root_item
);
806 ret
= btrfs_write_dirty_block_groups(trans
, root
);
811 if (root
!= root
->fs_info
->extent_root
)
812 switch_commit_root(root
);
818 * update all the cowonly tree roots on disk
820 * The error handling in this function may not be obvious. Any of the
821 * failures will cause the file system to go offline. We still need
822 * to clean up the delayed refs.
824 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
,
825 struct btrfs_root
*root
)
827 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
828 struct list_head
*next
;
829 struct extent_buffer
*eb
;
832 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
836 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
837 ret
= btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
,
839 btrfs_tree_unlock(eb
);
840 free_extent_buffer(eb
);
845 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
849 ret
= btrfs_run_dev_stats(trans
, root
->fs_info
);
851 ret
= btrfs_run_dev_replace(trans
, root
->fs_info
);
854 ret
= btrfs_run_qgroups(trans
, root
->fs_info
);
857 /* run_qgroups might have added some more refs */
858 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
861 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
862 next
= fs_info
->dirty_cowonly_roots
.next
;
864 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
866 ret
= update_cowonly_root(trans
, root
);
871 down_write(&fs_info
->extent_commit_sem
);
872 switch_commit_root(fs_info
->extent_root
);
873 up_write(&fs_info
->extent_commit_sem
);
875 btrfs_after_dev_replace_commit(fs_info
);
881 * dead roots are old snapshots that need to be deleted. This allocates
882 * a dirty root struct and adds it into the list of dead roots that need to
885 int btrfs_add_dead_root(struct btrfs_root
*root
)
887 spin_lock(&root
->fs_info
->trans_lock
);
888 list_add(&root
->root_list
, &root
->fs_info
->dead_roots
);
889 spin_unlock(&root
->fs_info
->trans_lock
);
894 * update all the cowonly tree roots on disk
896 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
,
897 struct btrfs_root
*root
)
899 struct btrfs_root
*gang
[8];
900 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
905 spin_lock(&fs_info
->fs_roots_radix_lock
);
907 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
910 BTRFS_ROOT_TRANS_TAG
);
913 for (i
= 0; i
< ret
; i
++) {
915 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
916 (unsigned long)root
->root_key
.objectid
,
917 BTRFS_ROOT_TRANS_TAG
);
918 spin_unlock(&fs_info
->fs_roots_radix_lock
);
920 btrfs_free_log(trans
, root
);
921 btrfs_update_reloc_root(trans
, root
);
922 btrfs_orphan_commit_root(trans
, root
);
924 btrfs_save_ino_cache(root
, trans
);
926 /* see comments in should_cow_block() */
930 if (root
->commit_root
!= root
->node
) {
931 mutex_lock(&root
->fs_commit_mutex
);
932 switch_commit_root(root
);
933 btrfs_unpin_free_ino(root
);
934 mutex_unlock(&root
->fs_commit_mutex
);
936 btrfs_set_root_node(&root
->root_item
,
940 err
= btrfs_update_root(trans
, fs_info
->tree_root
,
943 spin_lock(&fs_info
->fs_roots_radix_lock
);
948 spin_unlock(&fs_info
->fs_roots_radix_lock
);
953 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
954 * otherwise every leaf in the btree is read and defragged.
956 int btrfs_defrag_root(struct btrfs_root
*root
, int cacheonly
)
958 struct btrfs_fs_info
*info
= root
->fs_info
;
959 struct btrfs_trans_handle
*trans
;
962 if (xchg(&root
->defrag_running
, 1))
966 trans
= btrfs_start_transaction(root
, 0);
968 return PTR_ERR(trans
);
970 ret
= btrfs_defrag_leaves(trans
, root
, cacheonly
);
972 btrfs_end_transaction(trans
, root
);
973 btrfs_btree_balance_dirty(info
->tree_root
);
976 if (btrfs_fs_closing(root
->fs_info
) || ret
!= -EAGAIN
)
979 root
->defrag_running
= 0;
984 * new snapshots need to be created at a very specific time in the
985 * transaction commit. This does the actual creation
987 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
988 struct btrfs_fs_info
*fs_info
,
989 struct btrfs_pending_snapshot
*pending
)
991 struct btrfs_key key
;
992 struct btrfs_root_item
*new_root_item
;
993 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
994 struct btrfs_root
*root
= pending
->root
;
995 struct btrfs_root
*parent_root
;
996 struct btrfs_block_rsv
*rsv
;
997 struct inode
*parent_inode
;
998 struct btrfs_path
*path
;
999 struct btrfs_dir_item
*dir_item
;
1000 struct dentry
*parent
;
1001 struct dentry
*dentry
;
1002 struct extent_buffer
*tmp
;
1003 struct extent_buffer
*old
;
1004 struct timespec cur_time
= CURRENT_TIME
;
1012 path
= btrfs_alloc_path();
1014 ret
= pending
->error
= -ENOMEM
;
1015 goto path_alloc_fail
;
1018 new_root_item
= kmalloc(sizeof(*new_root_item
), GFP_NOFS
);
1019 if (!new_root_item
) {
1020 ret
= pending
->error
= -ENOMEM
;
1021 goto root_item_alloc_fail
;
1024 ret
= btrfs_find_free_objectid(tree_root
, &objectid
);
1026 pending
->error
= ret
;
1027 goto no_free_objectid
;
1030 btrfs_reloc_pre_snapshot(trans
, pending
, &to_reserve
);
1032 if (to_reserve
> 0) {
1033 ret
= btrfs_block_rsv_add(root
, &pending
->block_rsv
,
1035 BTRFS_RESERVE_NO_FLUSH
);
1037 pending
->error
= ret
;
1038 goto no_free_objectid
;
1042 ret
= btrfs_qgroup_inherit(trans
, fs_info
, root
->root_key
.objectid
,
1043 objectid
, pending
->inherit
);
1045 pending
->error
= ret
;
1046 goto no_free_objectid
;
1049 key
.objectid
= objectid
;
1050 key
.offset
= (u64
)-1;
1051 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1053 rsv
= trans
->block_rsv
;
1054 trans
->block_rsv
= &pending
->block_rsv
;
1056 dentry
= pending
->dentry
;
1057 parent
= dget_parent(dentry
);
1058 parent_inode
= parent
->d_inode
;
1059 parent_root
= BTRFS_I(parent_inode
)->root
;
1060 record_root_in_trans(trans
, parent_root
);
1063 * insert the directory item
1065 ret
= btrfs_set_inode_index(parent_inode
, &index
);
1066 BUG_ON(ret
); /* -ENOMEM */
1068 /* check if there is a file/dir which has the same name. */
1069 dir_item
= btrfs_lookup_dir_item(NULL
, parent_root
, path
,
1070 btrfs_ino(parent_inode
),
1071 dentry
->d_name
.name
,
1072 dentry
->d_name
.len
, 0);
1073 if (dir_item
!= NULL
&& !IS_ERR(dir_item
)) {
1074 pending
->error
= -EEXIST
;
1076 } else if (IS_ERR(dir_item
)) {
1077 ret
= PTR_ERR(dir_item
);
1078 btrfs_abort_transaction(trans
, root
, ret
);
1081 btrfs_release_path(path
);
1084 * pull in the delayed directory update
1085 * and the delayed inode item
1086 * otherwise we corrupt the FS during
1089 ret
= btrfs_run_delayed_items(trans
, root
);
1090 if (ret
) { /* Transaction aborted */
1091 btrfs_abort_transaction(trans
, root
, ret
);
1095 record_root_in_trans(trans
, root
);
1096 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
1097 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
1098 btrfs_check_and_init_root_item(new_root_item
);
1100 root_flags
= btrfs_root_flags(new_root_item
);
1101 if (pending
->readonly
)
1102 root_flags
|= BTRFS_ROOT_SUBVOL_RDONLY
;
1104 root_flags
&= ~BTRFS_ROOT_SUBVOL_RDONLY
;
1105 btrfs_set_root_flags(new_root_item
, root_flags
);
1107 btrfs_set_root_generation_v2(new_root_item
,
1109 uuid_le_gen(&new_uuid
);
1110 memcpy(new_root_item
->uuid
, new_uuid
.b
, BTRFS_UUID_SIZE
);
1111 memcpy(new_root_item
->parent_uuid
, root
->root_item
.uuid
,
1113 new_root_item
->otime
.sec
= cpu_to_le64(cur_time
.tv_sec
);
1114 new_root_item
->otime
.nsec
= cpu_to_le32(cur_time
.tv_nsec
);
1115 btrfs_set_root_otransid(new_root_item
, trans
->transid
);
1116 memset(&new_root_item
->stime
, 0, sizeof(new_root_item
->stime
));
1117 memset(&new_root_item
->rtime
, 0, sizeof(new_root_item
->rtime
));
1118 btrfs_set_root_stransid(new_root_item
, 0);
1119 btrfs_set_root_rtransid(new_root_item
, 0);
1121 old
= btrfs_lock_root_node(root
);
1122 ret
= btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
);
1124 btrfs_tree_unlock(old
);
1125 free_extent_buffer(old
);
1126 btrfs_abort_transaction(trans
, root
, ret
);
1130 btrfs_set_lock_blocking(old
);
1132 ret
= btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
1133 /* clean up in any case */
1134 btrfs_tree_unlock(old
);
1135 free_extent_buffer(old
);
1137 btrfs_abort_transaction(trans
, root
, ret
);
1141 /* see comments in should_cow_block() */
1142 root
->force_cow
= 1;
1145 btrfs_set_root_node(new_root_item
, tmp
);
1146 /* record when the snapshot was created in key.offset */
1147 key
.offset
= trans
->transid
;
1148 ret
= btrfs_insert_root(trans
, tree_root
, &key
, new_root_item
);
1149 btrfs_tree_unlock(tmp
);
1150 free_extent_buffer(tmp
);
1152 btrfs_abort_transaction(trans
, root
, ret
);
1157 * insert root back/forward references
1159 ret
= btrfs_add_root_ref(trans
, tree_root
, objectid
,
1160 parent_root
->root_key
.objectid
,
1161 btrfs_ino(parent_inode
), index
,
1162 dentry
->d_name
.name
, dentry
->d_name
.len
);
1164 btrfs_abort_transaction(trans
, root
, ret
);
1168 key
.offset
= (u64
)-1;
1169 pending
->snap
= btrfs_read_fs_root_no_name(root
->fs_info
, &key
);
1170 if (IS_ERR(pending
->snap
)) {
1171 ret
= PTR_ERR(pending
->snap
);
1172 btrfs_abort_transaction(trans
, root
, ret
);
1176 ret
= btrfs_reloc_post_snapshot(trans
, pending
);
1178 btrfs_abort_transaction(trans
, root
, ret
);
1182 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1184 btrfs_abort_transaction(trans
, root
, ret
);
1188 ret
= btrfs_insert_dir_item(trans
, parent_root
,
1189 dentry
->d_name
.name
, dentry
->d_name
.len
,
1191 BTRFS_FT_DIR
, index
);
1192 /* We have check then name at the beginning, so it is impossible. */
1193 BUG_ON(ret
== -EEXIST
|| ret
== -EOVERFLOW
);
1195 btrfs_abort_transaction(trans
, root
, ret
);
1199 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
1200 dentry
->d_name
.len
* 2);
1201 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
1202 ret
= btrfs_update_inode_fallback(trans
, parent_root
, parent_inode
);
1204 btrfs_abort_transaction(trans
, root
, ret
);
1207 trans
->block_rsv
= rsv
;
1209 kfree(new_root_item
);
1210 root_item_alloc_fail
:
1211 btrfs_free_path(path
);
1213 btrfs_block_rsv_release(root
, &pending
->block_rsv
, (u64
)-1);
1218 * create all the snapshots we've scheduled for creation
1220 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
,
1221 struct btrfs_fs_info
*fs_info
)
1223 struct btrfs_pending_snapshot
*pending
;
1224 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
1226 list_for_each_entry(pending
, head
, list
)
1227 create_pending_snapshot(trans
, fs_info
, pending
);
1231 static void update_super_roots(struct btrfs_root
*root
)
1233 struct btrfs_root_item
*root_item
;
1234 struct btrfs_super_block
*super
;
1236 super
= root
->fs_info
->super_copy
;
1238 root_item
= &root
->fs_info
->chunk_root
->root_item
;
1239 super
->chunk_root
= root_item
->bytenr
;
1240 super
->chunk_root_generation
= root_item
->generation
;
1241 super
->chunk_root_level
= root_item
->level
;
1243 root_item
= &root
->fs_info
->tree_root
->root_item
;
1244 super
->root
= root_item
->bytenr
;
1245 super
->generation
= root_item
->generation
;
1246 super
->root_level
= root_item
->level
;
1247 if (btrfs_test_opt(root
, SPACE_CACHE
))
1248 super
->cache_generation
= root_item
->generation
;
1251 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
1254 spin_lock(&info
->trans_lock
);
1255 if (info
->running_transaction
)
1256 ret
= info
->running_transaction
->in_commit
;
1257 spin_unlock(&info
->trans_lock
);
1261 int btrfs_transaction_blocked(struct btrfs_fs_info
*info
)
1264 spin_lock(&info
->trans_lock
);
1265 if (info
->running_transaction
)
1266 ret
= info
->running_transaction
->blocked
;
1267 spin_unlock(&info
->trans_lock
);
1272 * wait for the current transaction commit to start and block subsequent
1275 static void wait_current_trans_commit_start(struct btrfs_root
*root
,
1276 struct btrfs_transaction
*trans
)
1278 wait_event(root
->fs_info
->transaction_blocked_wait
, trans
->in_commit
);
1282 * wait for the current transaction to start and then become unblocked.
1285 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root
*root
,
1286 struct btrfs_transaction
*trans
)
1288 wait_event(root
->fs_info
->transaction_wait
,
1289 trans
->commit_done
|| (trans
->in_commit
&& !trans
->blocked
));
1293 * commit transactions asynchronously. once btrfs_commit_transaction_async
1294 * returns, any subsequent transaction will not be allowed to join.
1296 struct btrfs_async_commit
{
1297 struct btrfs_trans_handle
*newtrans
;
1298 struct btrfs_root
*root
;
1299 struct delayed_work work
;
1302 static void do_async_commit(struct work_struct
*work
)
1304 struct btrfs_async_commit
*ac
=
1305 container_of(work
, struct btrfs_async_commit
, work
.work
);
1308 * We've got freeze protection passed with the transaction.
1309 * Tell lockdep about it.
1311 if (ac
->newtrans
->type
< TRANS_JOIN_NOLOCK
)
1313 &ac
->root
->fs_info
->sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
1316 current
->journal_info
= ac
->newtrans
;
1318 btrfs_commit_transaction(ac
->newtrans
, ac
->root
);
1322 int btrfs_commit_transaction_async(struct btrfs_trans_handle
*trans
,
1323 struct btrfs_root
*root
,
1324 int wait_for_unblock
)
1326 struct btrfs_async_commit
*ac
;
1327 struct btrfs_transaction
*cur_trans
;
1329 ac
= kmalloc(sizeof(*ac
), GFP_NOFS
);
1333 INIT_DELAYED_WORK(&ac
->work
, do_async_commit
);
1335 ac
->newtrans
= btrfs_join_transaction(root
);
1336 if (IS_ERR(ac
->newtrans
)) {
1337 int err
= PTR_ERR(ac
->newtrans
);
1342 /* take transaction reference */
1343 cur_trans
= trans
->transaction
;
1344 atomic_inc(&cur_trans
->use_count
);
1346 btrfs_end_transaction(trans
, root
);
1349 * Tell lockdep we've released the freeze rwsem, since the
1350 * async commit thread will be the one to unlock it.
1352 if (trans
->type
< TRANS_JOIN_NOLOCK
)
1354 &root
->fs_info
->sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
1357 schedule_delayed_work(&ac
->work
, 0);
1359 /* wait for transaction to start and unblock */
1360 if (wait_for_unblock
)
1361 wait_current_trans_commit_start_and_unblock(root
, cur_trans
);
1363 wait_current_trans_commit_start(root
, cur_trans
);
1365 if (current
->journal_info
== trans
)
1366 current
->journal_info
= NULL
;
1368 put_transaction(cur_trans
);
1373 static void cleanup_transaction(struct btrfs_trans_handle
*trans
,
1374 struct btrfs_root
*root
, int err
)
1376 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1378 WARN_ON(trans
->use_count
> 1);
1380 btrfs_abort_transaction(trans
, root
, err
);
1382 spin_lock(&root
->fs_info
->trans_lock
);
1383 list_del_init(&cur_trans
->list
);
1384 if (cur_trans
== root
->fs_info
->running_transaction
) {
1385 root
->fs_info
->running_transaction
= NULL
;
1386 root
->fs_info
->trans_no_join
= 0;
1388 spin_unlock(&root
->fs_info
->trans_lock
);
1390 btrfs_cleanup_one_transaction(trans
->transaction
, root
);
1392 put_transaction(cur_trans
);
1393 put_transaction(cur_trans
);
1395 trace_btrfs_transaction_commit(root
);
1397 btrfs_scrub_continue(root
);
1399 if (current
->journal_info
== trans
)
1400 current
->journal_info
= NULL
;
1402 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1405 static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle
*trans
,
1406 struct btrfs_root
*root
)
1408 int flush_on_commit
= btrfs_test_opt(root
, FLUSHONCOMMIT
);
1409 int snap_pending
= 0;
1412 if (!flush_on_commit
) {
1413 spin_lock(&root
->fs_info
->trans_lock
);
1414 if (!list_empty(&trans
->transaction
->pending_snapshots
))
1416 spin_unlock(&root
->fs_info
->trans_lock
);
1419 if (flush_on_commit
|| snap_pending
) {
1420 btrfs_start_delalloc_inodes(root
, 1);
1421 btrfs_wait_ordered_extents(root
, 1);
1424 ret
= btrfs_run_delayed_items(trans
, root
);
1429 * running the delayed items may have added new refs. account
1430 * them now so that they hinder processing of more delayed refs
1431 * as little as possible.
1433 btrfs_delayed_refs_qgroup_accounting(trans
, root
->fs_info
);
1436 * rename don't use btrfs_join_transaction, so, once we
1437 * set the transaction to blocked above, we aren't going
1438 * to get any new ordered operations. We can safely run
1439 * it here and no for sure that nothing new will be added
1442 btrfs_run_ordered_operations(root
, 1);
1448 * btrfs_transaction state sequence:
1449 * in_commit = 0, blocked = 0 (initial)
1450 * in_commit = 1, blocked = 1
1454 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
,
1455 struct btrfs_root
*root
)
1457 unsigned long joined
= 0;
1458 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1459 struct btrfs_transaction
*prev_trans
= NULL
;
1462 int should_grow
= 0;
1463 unsigned long now
= get_seconds();
1465 ret
= btrfs_run_ordered_operations(root
, 0);
1467 btrfs_abort_transaction(trans
, root
, ret
);
1468 goto cleanup_transaction
;
1471 /* Stop the commit early if ->aborted is set */
1472 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1473 ret
= cur_trans
->aborted
;
1474 goto cleanup_transaction
;
1477 /* make a pass through all the delayed refs we have so far
1478 * any runnings procs may add more while we are here
1480 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1482 goto cleanup_transaction
;
1484 btrfs_trans_release_metadata(trans
, root
);
1485 trans
->block_rsv
= NULL
;
1487 cur_trans
= trans
->transaction
;
1490 * set the flushing flag so procs in this transaction have to
1491 * start sending their work down.
1493 cur_trans
->delayed_refs
.flushing
= 1;
1495 if (!list_empty(&trans
->new_bgs
))
1496 btrfs_create_pending_block_groups(trans
, root
);
1498 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1500 goto cleanup_transaction
;
1502 spin_lock(&cur_trans
->commit_lock
);
1503 if (cur_trans
->in_commit
) {
1504 spin_unlock(&cur_trans
->commit_lock
);
1505 atomic_inc(&cur_trans
->use_count
);
1506 ret
= btrfs_end_transaction(trans
, root
);
1508 wait_for_commit(root
, cur_trans
);
1510 put_transaction(cur_trans
);
1515 trans
->transaction
->in_commit
= 1;
1516 trans
->transaction
->blocked
= 1;
1517 spin_unlock(&cur_trans
->commit_lock
);
1518 wake_up(&root
->fs_info
->transaction_blocked_wait
);
1520 spin_lock(&root
->fs_info
->trans_lock
);
1521 if (cur_trans
->list
.prev
!= &root
->fs_info
->trans_list
) {
1522 prev_trans
= list_entry(cur_trans
->list
.prev
,
1523 struct btrfs_transaction
, list
);
1524 if (!prev_trans
->commit_done
) {
1525 atomic_inc(&prev_trans
->use_count
);
1526 spin_unlock(&root
->fs_info
->trans_lock
);
1528 wait_for_commit(root
, prev_trans
);
1530 put_transaction(prev_trans
);
1532 spin_unlock(&root
->fs_info
->trans_lock
);
1535 spin_unlock(&root
->fs_info
->trans_lock
);
1538 if (!btrfs_test_opt(root
, SSD
) &&
1539 (now
< cur_trans
->start_time
|| now
- cur_trans
->start_time
< 1))
1543 joined
= cur_trans
->num_joined
;
1545 WARN_ON(cur_trans
!= trans
->transaction
);
1547 ret
= btrfs_flush_all_pending_stuffs(trans
, root
);
1549 goto cleanup_transaction
;
1551 prepare_to_wait(&cur_trans
->writer_wait
, &wait
,
1552 TASK_UNINTERRUPTIBLE
);
1554 if (atomic_read(&cur_trans
->num_writers
) > 1)
1555 schedule_timeout(MAX_SCHEDULE_TIMEOUT
);
1556 else if (should_grow
)
1557 schedule_timeout(1);
1559 finish_wait(&cur_trans
->writer_wait
, &wait
);
1560 } while (atomic_read(&cur_trans
->num_writers
) > 1 ||
1561 (should_grow
&& cur_trans
->num_joined
!= joined
));
1563 ret
= btrfs_flush_all_pending_stuffs(trans
, root
);
1565 goto cleanup_transaction
;
1568 * Ok now we need to make sure to block out any other joins while we
1569 * commit the transaction. We could have started a join before setting
1570 * no_join so make sure to wait for num_writers to == 1 again.
1572 spin_lock(&root
->fs_info
->trans_lock
);
1573 root
->fs_info
->trans_no_join
= 1;
1574 spin_unlock(&root
->fs_info
->trans_lock
);
1575 wait_event(cur_trans
->writer_wait
,
1576 atomic_read(&cur_trans
->num_writers
) == 1);
1578 /* ->aborted might be set after the previous check, so check it */
1579 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1580 ret
= cur_trans
->aborted
;
1581 goto cleanup_transaction
;
1584 * the reloc mutex makes sure that we stop
1585 * the balancing code from coming in and moving
1586 * extents around in the middle of the commit
1588 mutex_lock(&root
->fs_info
->reloc_mutex
);
1591 * We needn't worry about the delayed items because we will
1592 * deal with them in create_pending_snapshot(), which is the
1593 * core function of the snapshot creation.
1595 ret
= create_pending_snapshots(trans
, root
->fs_info
);
1597 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1598 goto cleanup_transaction
;
1602 * We insert the dir indexes of the snapshots and update the inode
1603 * of the snapshots' parents after the snapshot creation, so there
1604 * are some delayed items which are not dealt with. Now deal with
1607 * We needn't worry that this operation will corrupt the snapshots,
1608 * because all the tree which are snapshoted will be forced to COW
1609 * the nodes and leaves.
1611 ret
= btrfs_run_delayed_items(trans
, root
);
1613 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1614 goto cleanup_transaction
;
1617 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1619 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1620 goto cleanup_transaction
;
1624 * make sure none of the code above managed to slip in a
1627 btrfs_assert_delayed_root_empty(root
);
1629 WARN_ON(cur_trans
!= trans
->transaction
);
1631 btrfs_scrub_pause(root
);
1632 /* btrfs_commit_tree_roots is responsible for getting the
1633 * various roots consistent with each other. Every pointer
1634 * in the tree of tree roots has to point to the most up to date
1635 * root for every subvolume and other tree. So, we have to keep
1636 * the tree logging code from jumping in and changing any
1639 * At this point in the commit, there can't be any tree-log
1640 * writers, but a little lower down we drop the trans mutex
1641 * and let new people in. By holding the tree_log_mutex
1642 * from now until after the super is written, we avoid races
1643 * with the tree-log code.
1645 mutex_lock(&root
->fs_info
->tree_log_mutex
);
1647 ret
= commit_fs_roots(trans
, root
);
1649 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1650 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1651 goto cleanup_transaction
;
1654 /* commit_fs_roots gets rid of all the tree log roots, it is now
1655 * safe to free the root of tree log roots
1657 btrfs_free_log_root_tree(trans
, root
->fs_info
);
1659 ret
= commit_cowonly_roots(trans
, root
);
1661 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1662 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1663 goto cleanup_transaction
;
1667 * The tasks which save the space cache and inode cache may also
1668 * update ->aborted, check it.
1670 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1671 ret
= cur_trans
->aborted
;
1672 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1673 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1674 goto cleanup_transaction
;
1677 btrfs_prepare_extent_commit(trans
, root
);
1679 cur_trans
= root
->fs_info
->running_transaction
;
1681 btrfs_set_root_node(&root
->fs_info
->tree_root
->root_item
,
1682 root
->fs_info
->tree_root
->node
);
1683 switch_commit_root(root
->fs_info
->tree_root
);
1685 btrfs_set_root_node(&root
->fs_info
->chunk_root
->root_item
,
1686 root
->fs_info
->chunk_root
->node
);
1687 switch_commit_root(root
->fs_info
->chunk_root
);
1689 assert_qgroups_uptodate(trans
);
1690 update_super_roots(root
);
1692 if (!root
->fs_info
->log_root_recovering
) {
1693 btrfs_set_super_log_root(root
->fs_info
->super_copy
, 0);
1694 btrfs_set_super_log_root_level(root
->fs_info
->super_copy
, 0);
1697 memcpy(root
->fs_info
->super_for_commit
, root
->fs_info
->super_copy
,
1698 sizeof(*root
->fs_info
->super_copy
));
1700 trans
->transaction
->blocked
= 0;
1701 spin_lock(&root
->fs_info
->trans_lock
);
1702 root
->fs_info
->running_transaction
= NULL
;
1703 root
->fs_info
->trans_no_join
= 0;
1704 spin_unlock(&root
->fs_info
->trans_lock
);
1705 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1707 wake_up(&root
->fs_info
->transaction_wait
);
1709 ret
= btrfs_write_and_wait_transaction(trans
, root
);
1711 btrfs_error(root
->fs_info
, ret
,
1712 "Error while writing out transaction.");
1713 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1714 goto cleanup_transaction
;
1717 ret
= write_ctree_super(trans
, root
, 0);
1719 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1720 goto cleanup_transaction
;
1724 * the super is written, we can safely allow the tree-loggers
1725 * to go about their business
1727 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1729 btrfs_finish_extent_commit(trans
, root
);
1731 cur_trans
->commit_done
= 1;
1733 root
->fs_info
->last_trans_committed
= cur_trans
->transid
;
1735 wake_up(&cur_trans
->commit_wait
);
1737 spin_lock(&root
->fs_info
->trans_lock
);
1738 list_del_init(&cur_trans
->list
);
1739 spin_unlock(&root
->fs_info
->trans_lock
);
1741 put_transaction(cur_trans
);
1742 put_transaction(cur_trans
);
1744 if (trans
->type
< TRANS_JOIN_NOLOCK
)
1745 sb_end_intwrite(root
->fs_info
->sb
);
1747 trace_btrfs_transaction_commit(root
);
1749 btrfs_scrub_continue(root
);
1751 if (current
->journal_info
== trans
)
1752 current
->journal_info
= NULL
;
1754 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1756 if (current
!= root
->fs_info
->transaction_kthread
)
1757 btrfs_run_delayed_iputs(root
);
1761 cleanup_transaction
:
1762 btrfs_trans_release_metadata(trans
, root
);
1763 trans
->block_rsv
= NULL
;
1764 btrfs_printk(root
->fs_info
, "Skipping commit of aborted transaction.\n");
1766 if (current
->journal_info
== trans
)
1767 current
->journal_info
= NULL
;
1768 cleanup_transaction(trans
, root
, ret
);
1774 * interface function to delete all the snapshots we have scheduled for deletion
1776 int btrfs_clean_old_snapshots(struct btrfs_root
*root
)
1779 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1781 spin_lock(&fs_info
->trans_lock
);
1782 list_splice_init(&fs_info
->dead_roots
, &list
);
1783 spin_unlock(&fs_info
->trans_lock
);
1785 while (!list_empty(&list
)) {
1788 root
= list_entry(list
.next
, struct btrfs_root
, root_list
);
1789 list_del(&root
->root_list
);
1791 btrfs_kill_all_delayed_nodes(root
);
1793 if (btrfs_header_backref_rev(root
->node
) <
1794 BTRFS_MIXED_BACKREF_REV
)
1795 ret
= btrfs_drop_snapshot(root
, NULL
, 0, 0);
1797 ret
=btrfs_drop_snapshot(root
, NULL
, 1, 0);