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 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
);
53 static inline int can_join_transaction(struct btrfs_transaction
*trans
,
56 return !(trans
->in_commit
&&
58 type
!= TRANS_JOIN_NOLOCK
);
62 * either allocate a new transaction or hop into the existing one
64 static noinline
int join_transaction(struct btrfs_root
*root
, int type
)
66 struct btrfs_transaction
*cur_trans
;
67 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
69 spin_lock(&fs_info
->trans_lock
);
71 /* The file system has been taken offline. No new transactions. */
72 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
73 spin_unlock(&fs_info
->trans_lock
);
77 if (fs_info
->trans_no_join
) {
79 * If we are JOIN_NOLOCK we're already committing a current
80 * transaction, we just need a handle to deal with something
81 * when committing the transaction, such as inode cache and
82 * space cache. It is a special case.
84 if (type
!= TRANS_JOIN_NOLOCK
) {
85 spin_unlock(&fs_info
->trans_lock
);
90 cur_trans
= fs_info
->running_transaction
;
92 if (cur_trans
->aborted
) {
93 spin_unlock(&fs_info
->trans_lock
);
94 return cur_trans
->aborted
;
96 if (!can_join_transaction(cur_trans
, type
)) {
97 spin_unlock(&fs_info
->trans_lock
);
100 atomic_inc(&cur_trans
->use_count
);
101 atomic_inc(&cur_trans
->num_writers
);
102 cur_trans
->num_joined
++;
103 spin_unlock(&fs_info
->trans_lock
);
106 spin_unlock(&fs_info
->trans_lock
);
109 * If we are ATTACH, we just want to catch the current transaction,
110 * and commit it. If there is no transaction, just return ENOENT.
112 if (type
== TRANS_ATTACH
)
115 cur_trans
= kmem_cache_alloc(btrfs_transaction_cachep
, GFP_NOFS
);
119 spin_lock(&fs_info
->trans_lock
);
120 if (fs_info
->running_transaction
) {
122 * someone started a transaction after we unlocked. Make sure
123 * to redo the trans_no_join checks above
125 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
126 cur_trans
= fs_info
->running_transaction
;
128 } else if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
129 spin_unlock(&fs_info
->trans_lock
);
130 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
134 atomic_set(&cur_trans
->num_writers
, 1);
135 cur_trans
->num_joined
= 0;
136 init_waitqueue_head(&cur_trans
->writer_wait
);
137 init_waitqueue_head(&cur_trans
->commit_wait
);
138 cur_trans
->in_commit
= 0;
139 cur_trans
->blocked
= 0;
141 * One for this trans handle, one so it will live on until we
142 * commit the transaction.
144 atomic_set(&cur_trans
->use_count
, 2);
145 cur_trans
->commit_done
= 0;
146 cur_trans
->start_time
= get_seconds();
148 cur_trans
->delayed_refs
.root
= RB_ROOT
;
149 cur_trans
->delayed_refs
.num_entries
= 0;
150 cur_trans
->delayed_refs
.num_heads_ready
= 0;
151 cur_trans
->delayed_refs
.num_heads
= 0;
152 cur_trans
->delayed_refs
.flushing
= 0;
153 cur_trans
->delayed_refs
.run_delayed_start
= 0;
156 * although the tree mod log is per file system and not per transaction,
157 * the log must never go across transaction boundaries.
160 if (!list_empty(&fs_info
->tree_mod_seq_list
))
161 WARN(1, KERN_ERR
"btrfs: tree_mod_seq_list not empty when "
162 "creating a fresh transaction\n");
163 if (!RB_EMPTY_ROOT(&fs_info
->tree_mod_log
))
164 WARN(1, KERN_ERR
"btrfs: tree_mod_log rb tree not empty when "
165 "creating a fresh transaction\n");
166 atomic_set(&fs_info
->tree_mod_seq
, 0);
168 spin_lock_init(&cur_trans
->commit_lock
);
169 spin_lock_init(&cur_trans
->delayed_refs
.lock
);
171 INIT_LIST_HEAD(&cur_trans
->pending_snapshots
);
172 INIT_LIST_HEAD(&cur_trans
->ordered_operations
);
173 list_add_tail(&cur_trans
->list
, &fs_info
->trans_list
);
174 extent_io_tree_init(&cur_trans
->dirty_pages
,
175 fs_info
->btree_inode
->i_mapping
);
176 fs_info
->generation
++;
177 cur_trans
->transid
= fs_info
->generation
;
178 fs_info
->running_transaction
= cur_trans
;
179 cur_trans
->aborted
= 0;
180 spin_unlock(&fs_info
->trans_lock
);
186 * this does all the record keeping required to make sure that a reference
187 * counted root is properly recorded in a given transaction. This is required
188 * to make sure the old root from before we joined the transaction is deleted
189 * when the transaction commits
191 static int record_root_in_trans(struct btrfs_trans_handle
*trans
,
192 struct btrfs_root
*root
)
194 if (root
->ref_cows
&& root
->last_trans
< trans
->transid
) {
195 WARN_ON(root
== root
->fs_info
->extent_root
);
196 WARN_ON(root
->commit_root
!= root
->node
);
199 * see below for in_trans_setup usage rules
200 * we have the reloc mutex held now, so there
201 * is only one writer in this function
203 root
->in_trans_setup
= 1;
205 /* make sure readers find in_trans_setup before
206 * they find our root->last_trans update
210 spin_lock(&root
->fs_info
->fs_roots_radix_lock
);
211 if (root
->last_trans
== trans
->transid
) {
212 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
215 radix_tree_tag_set(&root
->fs_info
->fs_roots_radix
,
216 (unsigned long)root
->root_key
.objectid
,
217 BTRFS_ROOT_TRANS_TAG
);
218 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
219 root
->last_trans
= trans
->transid
;
221 /* this is pretty tricky. We don't want to
222 * take the relocation lock in btrfs_record_root_in_trans
223 * unless we're really doing the first setup for this root in
226 * Normally we'd use root->last_trans as a flag to decide
227 * if we want to take the expensive mutex.
229 * But, we have to set root->last_trans before we
230 * init the relocation root, otherwise, we trip over warnings
231 * in ctree.c. The solution used here is to flag ourselves
232 * with root->in_trans_setup. When this is 1, we're still
233 * fixing up the reloc trees and everyone must wait.
235 * When this is zero, they can trust root->last_trans and fly
236 * through btrfs_record_root_in_trans without having to take the
237 * lock. smp_wmb() makes sure that all the writes above are
238 * done before we pop in the zero below
240 btrfs_init_reloc_root(trans
, root
);
242 root
->in_trans_setup
= 0;
248 int btrfs_record_root_in_trans(struct btrfs_trans_handle
*trans
,
249 struct btrfs_root
*root
)
255 * see record_root_in_trans for comments about in_trans_setup usage
259 if (root
->last_trans
== trans
->transid
&&
260 !root
->in_trans_setup
)
263 mutex_lock(&root
->fs_info
->reloc_mutex
);
264 record_root_in_trans(trans
, root
);
265 mutex_unlock(&root
->fs_info
->reloc_mutex
);
270 /* wait for commit against the current transaction to become unblocked
271 * when this is done, it is safe to start a new transaction, but the current
272 * transaction might not be fully on disk.
274 static void wait_current_trans(struct btrfs_root
*root
)
276 struct btrfs_transaction
*cur_trans
;
278 spin_lock(&root
->fs_info
->trans_lock
);
279 cur_trans
= root
->fs_info
->running_transaction
;
280 if (cur_trans
&& cur_trans
->blocked
) {
281 atomic_inc(&cur_trans
->use_count
);
282 spin_unlock(&root
->fs_info
->trans_lock
);
284 wait_event(root
->fs_info
->transaction_wait
,
285 !cur_trans
->blocked
);
286 put_transaction(cur_trans
);
288 spin_unlock(&root
->fs_info
->trans_lock
);
292 static int may_wait_transaction(struct btrfs_root
*root
, int type
)
294 if (root
->fs_info
->log_root_recovering
)
297 if (type
== TRANS_USERSPACE
)
300 if (type
== TRANS_START
&&
301 !atomic_read(&root
->fs_info
->open_ioctl_trans
))
307 static struct btrfs_trans_handle
*
308 start_transaction(struct btrfs_root
*root
, u64 num_items
, int type
,
309 enum btrfs_reserve_flush_enum flush
)
311 struct btrfs_trans_handle
*h
;
312 struct btrfs_transaction
*cur_trans
;
315 u64 qgroup_reserved
= 0;
317 if (test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
318 return ERR_PTR(-EROFS
);
320 if (current
->journal_info
) {
321 WARN_ON(type
!= TRANS_JOIN
&& type
!= TRANS_JOIN_NOLOCK
);
322 h
= current
->journal_info
;
324 WARN_ON(h
->use_count
> 2);
325 h
->orig_rsv
= h
->block_rsv
;
331 * Do the reservation before we join the transaction so we can do all
332 * the appropriate flushing if need be.
334 if (num_items
> 0 && root
!= root
->fs_info
->chunk_root
) {
335 if (root
->fs_info
->quota_enabled
&&
336 is_fstree(root
->root_key
.objectid
)) {
337 qgroup_reserved
= num_items
* root
->leafsize
;
338 ret
= btrfs_qgroup_reserve(root
, qgroup_reserved
);
343 num_bytes
= btrfs_calc_trans_metadata_size(root
, num_items
);
344 ret
= btrfs_block_rsv_add(root
,
345 &root
->fs_info
->trans_block_rsv
,
351 h
= kmem_cache_alloc(btrfs_trans_handle_cachep
, GFP_NOFS
);
358 * If we are JOIN_NOLOCK we're already committing a transaction and
359 * waiting on this guy, so we don't need to do the sb_start_intwrite
360 * because we're already holding a ref. We need this because we could
361 * have raced in and did an fsync() on a file which can kick a commit
362 * and then we deadlock with somebody doing a freeze.
364 * If we are ATTACH, it means we just want to catch the current
365 * transaction and commit it, so we needn't do sb_start_intwrite().
367 if (type
< TRANS_JOIN_NOLOCK
)
368 sb_start_intwrite(root
->fs_info
->sb
);
370 if (may_wait_transaction(root
, type
))
371 wait_current_trans(root
);
374 ret
= join_transaction(root
, type
);
376 wait_current_trans(root
);
377 if (unlikely(type
== TRANS_ATTACH
))
380 } while (ret
== -EBUSY
);
383 /* We must get the transaction if we are JOIN_NOLOCK. */
384 BUG_ON(type
== TRANS_JOIN_NOLOCK
);
388 cur_trans
= root
->fs_info
->running_transaction
;
390 h
->transid
= cur_trans
->transid
;
391 h
->transaction
= cur_trans
;
393 h
->bytes_reserved
= 0;
395 h
->delayed_ref_updates
= 0;
401 h
->qgroup_reserved
= 0;
402 h
->delayed_ref_elem
.seq
= 0;
404 h
->allocating_chunk
= false;
405 INIT_LIST_HEAD(&h
->qgroup_ref_list
);
406 INIT_LIST_HEAD(&h
->new_bgs
);
409 if (cur_trans
->blocked
&& may_wait_transaction(root
, type
)) {
410 btrfs_commit_transaction(h
, root
);
415 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
416 h
->transid
, num_bytes
, 1);
417 h
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
418 h
->bytes_reserved
= num_bytes
;
420 h
->qgroup_reserved
= qgroup_reserved
;
423 btrfs_record_root_in_trans(h
, root
);
425 if (!current
->journal_info
&& type
!= TRANS_USERSPACE
)
426 current
->journal_info
= h
;
430 if (type
< TRANS_JOIN_NOLOCK
)
431 sb_end_intwrite(root
->fs_info
->sb
);
432 kmem_cache_free(btrfs_trans_handle_cachep
, h
);
435 btrfs_block_rsv_release(root
, &root
->fs_info
->trans_block_rsv
,
439 btrfs_qgroup_free(root
, qgroup_reserved
);
443 struct btrfs_trans_handle
*btrfs_start_transaction(struct btrfs_root
*root
,
446 return start_transaction(root
, num_items
, TRANS_START
,
447 BTRFS_RESERVE_FLUSH_ALL
);
450 struct btrfs_trans_handle
*btrfs_start_transaction_lflush(
451 struct btrfs_root
*root
, int num_items
)
453 return start_transaction(root
, num_items
, TRANS_START
,
454 BTRFS_RESERVE_FLUSH_LIMIT
);
457 struct btrfs_trans_handle
*btrfs_join_transaction(struct btrfs_root
*root
)
459 return start_transaction(root
, 0, TRANS_JOIN
, 0);
462 struct btrfs_trans_handle
*btrfs_join_transaction_nolock(struct btrfs_root
*root
)
464 return start_transaction(root
, 0, TRANS_JOIN_NOLOCK
, 0);
467 struct btrfs_trans_handle
*btrfs_start_ioctl_transaction(struct btrfs_root
*root
)
469 return start_transaction(root
, 0, TRANS_USERSPACE
, 0);
472 struct btrfs_trans_handle
*btrfs_attach_transaction(struct btrfs_root
*root
)
474 return start_transaction(root
, 0, TRANS_ATTACH
, 0);
477 /* wait for a transaction commit to be fully complete */
478 static noinline
void wait_for_commit(struct btrfs_root
*root
,
479 struct btrfs_transaction
*commit
)
481 wait_event(commit
->commit_wait
, commit
->commit_done
);
484 int btrfs_wait_for_commit(struct btrfs_root
*root
, u64 transid
)
486 struct btrfs_transaction
*cur_trans
= NULL
, *t
;
490 if (transid
<= root
->fs_info
->last_trans_committed
)
494 /* find specified transaction */
495 spin_lock(&root
->fs_info
->trans_lock
);
496 list_for_each_entry(t
, &root
->fs_info
->trans_list
, list
) {
497 if (t
->transid
== transid
) {
499 atomic_inc(&cur_trans
->use_count
);
503 if (t
->transid
> transid
) {
508 spin_unlock(&root
->fs_info
->trans_lock
);
509 /* The specified transaction doesn't exist */
513 /* find newest transaction that is committing | committed */
514 spin_lock(&root
->fs_info
->trans_lock
);
515 list_for_each_entry_reverse(t
, &root
->fs_info
->trans_list
,
521 atomic_inc(&cur_trans
->use_count
);
525 spin_unlock(&root
->fs_info
->trans_lock
);
527 goto out
; /* nothing committing|committed */
530 wait_for_commit(root
, cur_trans
);
531 put_transaction(cur_trans
);
536 void btrfs_throttle(struct btrfs_root
*root
)
538 if (!atomic_read(&root
->fs_info
->open_ioctl_trans
))
539 wait_current_trans(root
);
542 static int should_end_transaction(struct btrfs_trans_handle
*trans
,
543 struct btrfs_root
*root
)
547 ret
= btrfs_block_rsv_check(root
, &root
->fs_info
->global_block_rsv
, 5);
551 int btrfs_should_end_transaction(struct btrfs_trans_handle
*trans
,
552 struct btrfs_root
*root
)
554 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
559 if (cur_trans
->blocked
|| cur_trans
->delayed_refs
.flushing
)
562 updates
= trans
->delayed_ref_updates
;
563 trans
->delayed_ref_updates
= 0;
565 err
= btrfs_run_delayed_refs(trans
, root
, updates
);
566 if (err
) /* Error code will also eval true */
570 return should_end_transaction(trans
, root
);
573 static int __btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
574 struct btrfs_root
*root
, int throttle
)
576 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
577 struct btrfs_fs_info
*info
= root
->fs_info
;
579 int lock
= (trans
->type
!= TRANS_JOIN_NOLOCK
);
582 if (--trans
->use_count
) {
583 trans
->block_rsv
= trans
->orig_rsv
;
588 * do the qgroup accounting as early as possible
590 err
= btrfs_delayed_refs_qgroup_accounting(trans
, info
);
592 btrfs_trans_release_metadata(trans
, root
);
593 trans
->block_rsv
= NULL
;
595 * the same root has to be passed to start_transaction and
596 * end_transaction. Subvolume quota depends on this.
598 WARN_ON(trans
->root
!= root
);
600 if (trans
->qgroup_reserved
) {
601 btrfs_qgroup_free(root
, trans
->qgroup_reserved
);
602 trans
->qgroup_reserved
= 0;
605 if (!list_empty(&trans
->new_bgs
))
606 btrfs_create_pending_block_groups(trans
, root
);
609 unsigned long cur
= trans
->delayed_ref_updates
;
610 trans
->delayed_ref_updates
= 0;
612 trans
->transaction
->delayed_refs
.num_heads_ready
> 64) {
613 trans
->delayed_ref_updates
= 0;
614 btrfs_run_delayed_refs(trans
, root
, cur
);
620 btrfs_trans_release_metadata(trans
, root
);
621 trans
->block_rsv
= NULL
;
623 if (!list_empty(&trans
->new_bgs
))
624 btrfs_create_pending_block_groups(trans
, root
);
626 if (lock
&& !atomic_read(&root
->fs_info
->open_ioctl_trans
) &&
627 should_end_transaction(trans
, root
)) {
628 trans
->transaction
->blocked
= 1;
632 if (lock
&& cur_trans
->blocked
&& !cur_trans
->in_commit
) {
635 * We may race with somebody else here so end up having
636 * to call end_transaction on ourselves again, so inc
640 return btrfs_commit_transaction(trans
, root
);
642 wake_up_process(info
->transaction_kthread
);
646 if (trans
->type
< TRANS_JOIN_NOLOCK
)
647 sb_end_intwrite(root
->fs_info
->sb
);
649 WARN_ON(cur_trans
!= info
->running_transaction
);
650 WARN_ON(atomic_read(&cur_trans
->num_writers
) < 1);
651 atomic_dec(&cur_trans
->num_writers
);
654 if (waitqueue_active(&cur_trans
->writer_wait
))
655 wake_up(&cur_trans
->writer_wait
);
656 put_transaction(cur_trans
);
658 if (current
->journal_info
== trans
)
659 current
->journal_info
= NULL
;
662 btrfs_run_delayed_iputs(root
);
664 if (trans
->aborted
||
665 test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
667 assert_qgroups_uptodate(trans
);
669 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
673 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
674 struct btrfs_root
*root
)
678 ret
= __btrfs_end_transaction(trans
, root
, 0);
684 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
,
685 struct btrfs_root
*root
)
689 ret
= __btrfs_end_transaction(trans
, root
, 1);
695 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle
*trans
,
696 struct btrfs_root
*root
)
698 return __btrfs_end_transaction(trans
, root
, 1);
702 * when btree blocks are allocated, they have some corresponding bits set for
703 * them in one of two extent_io trees. This is used to make sure all of
704 * those extents are sent to disk but does not wait on them
706 int btrfs_write_marked_extents(struct btrfs_root
*root
,
707 struct extent_io_tree
*dirty_pages
, int mark
)
711 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
712 struct extent_state
*cached_state
= NULL
;
716 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
717 mark
, &cached_state
)) {
718 convert_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
,
719 mark
, &cached_state
, GFP_NOFS
);
721 err
= filemap_fdatawrite_range(mapping
, start
, end
);
733 * when btree blocks are allocated, they have some corresponding bits set for
734 * them in one of two extent_io trees. This is used to make sure all of
735 * those extents are on disk for transaction or log commit. We wait
736 * on all the pages and clear them from the dirty pages state tree
738 int btrfs_wait_marked_extents(struct btrfs_root
*root
,
739 struct extent_io_tree
*dirty_pages
, int mark
)
743 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
744 struct extent_state
*cached_state
= NULL
;
748 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
749 EXTENT_NEED_WAIT
, &cached_state
)) {
750 clear_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
,
751 0, 0, &cached_state
, GFP_NOFS
);
752 err
= filemap_fdatawait_range(mapping
, start
, end
);
764 * when btree blocks are allocated, they have some corresponding bits set for
765 * them in one of two extent_io trees. This is used to make sure all of
766 * those extents are on disk for transaction or log commit
768 int btrfs_write_and_wait_marked_extents(struct btrfs_root
*root
,
769 struct extent_io_tree
*dirty_pages
, int mark
)
774 ret
= btrfs_write_marked_extents(root
, dirty_pages
, mark
);
775 ret2
= btrfs_wait_marked_extents(root
, dirty_pages
, mark
);
784 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
,
785 struct btrfs_root
*root
)
787 if (!trans
|| !trans
->transaction
) {
788 struct inode
*btree_inode
;
789 btree_inode
= root
->fs_info
->btree_inode
;
790 return filemap_write_and_wait(btree_inode
->i_mapping
);
792 return btrfs_write_and_wait_marked_extents(root
,
793 &trans
->transaction
->dirty_pages
,
798 * this is used to update the root pointer in the tree of tree roots.
800 * But, in the case of the extent allocation tree, updating the root
801 * pointer may allocate blocks which may change the root of the extent
804 * So, this loops and repeats and makes sure the cowonly root didn't
805 * change while the root pointer was being updated in the metadata.
807 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
808 struct btrfs_root
*root
)
813 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
815 old_root_used
= btrfs_root_used(&root
->root_item
);
816 btrfs_write_dirty_block_groups(trans
, root
);
819 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
820 if (old_root_bytenr
== root
->node
->start
&&
821 old_root_used
== btrfs_root_used(&root
->root_item
))
824 btrfs_set_root_node(&root
->root_item
, root
->node
);
825 ret
= btrfs_update_root(trans
, tree_root
,
831 old_root_used
= btrfs_root_used(&root
->root_item
);
832 ret
= btrfs_write_dirty_block_groups(trans
, root
);
837 if (root
!= root
->fs_info
->extent_root
)
838 switch_commit_root(root
);
844 * update all the cowonly tree roots on disk
846 * The error handling in this function may not be obvious. Any of the
847 * failures will cause the file system to go offline. We still need
848 * to clean up the delayed refs.
850 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
,
851 struct btrfs_root
*root
)
853 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
854 struct list_head
*next
;
855 struct extent_buffer
*eb
;
858 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
862 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
863 ret
= btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
,
865 btrfs_tree_unlock(eb
);
866 free_extent_buffer(eb
);
871 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
875 ret
= btrfs_run_dev_stats(trans
, root
->fs_info
);
877 ret
= btrfs_run_dev_replace(trans
, root
->fs_info
);
880 ret
= btrfs_run_qgroups(trans
, root
->fs_info
);
883 /* run_qgroups might have added some more refs */
884 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
887 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
888 next
= fs_info
->dirty_cowonly_roots
.next
;
890 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
892 ret
= update_cowonly_root(trans
, root
);
897 down_write(&fs_info
->extent_commit_sem
);
898 switch_commit_root(fs_info
->extent_root
);
899 up_write(&fs_info
->extent_commit_sem
);
901 btrfs_after_dev_replace_commit(fs_info
);
907 * dead roots are old snapshots that need to be deleted. This allocates
908 * a dirty root struct and adds it into the list of dead roots that need to
911 int btrfs_add_dead_root(struct btrfs_root
*root
)
913 spin_lock(&root
->fs_info
->trans_lock
);
914 list_add(&root
->root_list
, &root
->fs_info
->dead_roots
);
915 spin_unlock(&root
->fs_info
->trans_lock
);
920 * update all the cowonly tree roots on disk
922 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
,
923 struct btrfs_root
*root
)
925 struct btrfs_root
*gang
[8];
926 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
931 spin_lock(&fs_info
->fs_roots_radix_lock
);
933 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
936 BTRFS_ROOT_TRANS_TAG
);
939 for (i
= 0; i
< ret
; i
++) {
941 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
942 (unsigned long)root
->root_key
.objectid
,
943 BTRFS_ROOT_TRANS_TAG
);
944 spin_unlock(&fs_info
->fs_roots_radix_lock
);
946 btrfs_free_log(trans
, root
);
947 btrfs_update_reloc_root(trans
, root
);
948 btrfs_orphan_commit_root(trans
, root
);
950 btrfs_save_ino_cache(root
, trans
);
952 /* see comments in should_cow_block() */
956 if (root
->commit_root
!= root
->node
) {
957 mutex_lock(&root
->fs_commit_mutex
);
958 switch_commit_root(root
);
959 btrfs_unpin_free_ino(root
);
960 mutex_unlock(&root
->fs_commit_mutex
);
962 btrfs_set_root_node(&root
->root_item
,
966 err
= btrfs_update_root(trans
, fs_info
->tree_root
,
969 spin_lock(&fs_info
->fs_roots_radix_lock
);
974 spin_unlock(&fs_info
->fs_roots_radix_lock
);
979 * defrag a given btree.
980 * Every leaf in the btree is read and defragged.
982 int btrfs_defrag_root(struct btrfs_root
*root
)
984 struct btrfs_fs_info
*info
= root
->fs_info
;
985 struct btrfs_trans_handle
*trans
;
988 if (xchg(&root
->defrag_running
, 1))
992 trans
= btrfs_start_transaction(root
, 0);
994 return PTR_ERR(trans
);
996 ret
= btrfs_defrag_leaves(trans
, root
);
998 btrfs_end_transaction(trans
, root
);
999 btrfs_btree_balance_dirty(info
->tree_root
);
1002 if (btrfs_fs_closing(root
->fs_info
) || ret
!= -EAGAIN
)
1005 if (btrfs_defrag_cancelled(root
->fs_info
)) {
1006 printk(KERN_DEBUG
"btrfs: defrag_root cancelled\n");
1011 root
->defrag_running
= 0;
1016 * new snapshots need to be created at a very specific time in the
1017 * transaction commit. This does the actual creation
1019 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
1020 struct btrfs_fs_info
*fs_info
,
1021 struct btrfs_pending_snapshot
*pending
)
1023 struct btrfs_key key
;
1024 struct btrfs_root_item
*new_root_item
;
1025 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1026 struct btrfs_root
*root
= pending
->root
;
1027 struct btrfs_root
*parent_root
;
1028 struct btrfs_block_rsv
*rsv
;
1029 struct inode
*parent_inode
;
1030 struct btrfs_path
*path
;
1031 struct btrfs_dir_item
*dir_item
;
1032 struct dentry
*parent
;
1033 struct dentry
*dentry
;
1034 struct extent_buffer
*tmp
;
1035 struct extent_buffer
*old
;
1036 struct timespec cur_time
= CURRENT_TIME
;
1044 path
= btrfs_alloc_path();
1046 ret
= pending
->error
= -ENOMEM
;
1047 goto path_alloc_fail
;
1050 new_root_item
= kmalloc(sizeof(*new_root_item
), GFP_NOFS
);
1051 if (!new_root_item
) {
1052 ret
= pending
->error
= -ENOMEM
;
1053 goto root_item_alloc_fail
;
1056 ret
= btrfs_find_free_objectid(tree_root
, &objectid
);
1058 pending
->error
= ret
;
1059 goto no_free_objectid
;
1062 btrfs_reloc_pre_snapshot(trans
, pending
, &to_reserve
);
1064 if (to_reserve
> 0) {
1065 ret
= btrfs_block_rsv_add(root
, &pending
->block_rsv
,
1067 BTRFS_RESERVE_NO_FLUSH
);
1069 pending
->error
= ret
;
1070 goto no_free_objectid
;
1074 ret
= btrfs_qgroup_inherit(trans
, fs_info
, root
->root_key
.objectid
,
1075 objectid
, pending
->inherit
);
1077 pending
->error
= ret
;
1078 goto no_free_objectid
;
1081 key
.objectid
= objectid
;
1082 key
.offset
= (u64
)-1;
1083 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1085 rsv
= trans
->block_rsv
;
1086 trans
->block_rsv
= &pending
->block_rsv
;
1088 dentry
= pending
->dentry
;
1089 parent
= dget_parent(dentry
);
1090 parent_inode
= parent
->d_inode
;
1091 parent_root
= BTRFS_I(parent_inode
)->root
;
1092 record_root_in_trans(trans
, parent_root
);
1095 * insert the directory item
1097 ret
= btrfs_set_inode_index(parent_inode
, &index
);
1098 BUG_ON(ret
); /* -ENOMEM */
1100 /* check if there is a file/dir which has the same name. */
1101 dir_item
= btrfs_lookup_dir_item(NULL
, parent_root
, path
,
1102 btrfs_ino(parent_inode
),
1103 dentry
->d_name
.name
,
1104 dentry
->d_name
.len
, 0);
1105 if (dir_item
!= NULL
&& !IS_ERR(dir_item
)) {
1106 pending
->error
= -EEXIST
;
1108 } else if (IS_ERR(dir_item
)) {
1109 ret
= PTR_ERR(dir_item
);
1110 btrfs_abort_transaction(trans
, root
, ret
);
1113 btrfs_release_path(path
);
1116 * pull in the delayed directory update
1117 * and the delayed inode item
1118 * otherwise we corrupt the FS during
1121 ret
= btrfs_run_delayed_items(trans
, root
);
1122 if (ret
) { /* Transaction aborted */
1123 btrfs_abort_transaction(trans
, root
, ret
);
1127 record_root_in_trans(trans
, root
);
1128 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
1129 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
1130 btrfs_check_and_init_root_item(new_root_item
);
1132 root_flags
= btrfs_root_flags(new_root_item
);
1133 if (pending
->readonly
)
1134 root_flags
|= BTRFS_ROOT_SUBVOL_RDONLY
;
1136 root_flags
&= ~BTRFS_ROOT_SUBVOL_RDONLY
;
1137 btrfs_set_root_flags(new_root_item
, root_flags
);
1139 btrfs_set_root_generation_v2(new_root_item
,
1141 uuid_le_gen(&new_uuid
);
1142 memcpy(new_root_item
->uuid
, new_uuid
.b
, BTRFS_UUID_SIZE
);
1143 memcpy(new_root_item
->parent_uuid
, root
->root_item
.uuid
,
1145 new_root_item
->otime
.sec
= cpu_to_le64(cur_time
.tv_sec
);
1146 new_root_item
->otime
.nsec
= cpu_to_le32(cur_time
.tv_nsec
);
1147 btrfs_set_root_otransid(new_root_item
, trans
->transid
);
1148 memset(&new_root_item
->stime
, 0, sizeof(new_root_item
->stime
));
1149 memset(&new_root_item
->rtime
, 0, sizeof(new_root_item
->rtime
));
1150 btrfs_set_root_stransid(new_root_item
, 0);
1151 btrfs_set_root_rtransid(new_root_item
, 0);
1153 old
= btrfs_lock_root_node(root
);
1154 ret
= btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
);
1156 btrfs_tree_unlock(old
);
1157 free_extent_buffer(old
);
1158 btrfs_abort_transaction(trans
, root
, ret
);
1162 btrfs_set_lock_blocking(old
);
1164 ret
= btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
1165 /* clean up in any case */
1166 btrfs_tree_unlock(old
);
1167 free_extent_buffer(old
);
1169 btrfs_abort_transaction(trans
, root
, ret
);
1173 /* see comments in should_cow_block() */
1174 root
->force_cow
= 1;
1177 btrfs_set_root_node(new_root_item
, tmp
);
1178 /* record when the snapshot was created in key.offset */
1179 key
.offset
= trans
->transid
;
1180 ret
= btrfs_insert_root(trans
, tree_root
, &key
, new_root_item
);
1181 btrfs_tree_unlock(tmp
);
1182 free_extent_buffer(tmp
);
1184 btrfs_abort_transaction(trans
, root
, ret
);
1189 * insert root back/forward references
1191 ret
= btrfs_add_root_ref(trans
, tree_root
, objectid
,
1192 parent_root
->root_key
.objectid
,
1193 btrfs_ino(parent_inode
), index
,
1194 dentry
->d_name
.name
, dentry
->d_name
.len
);
1196 btrfs_abort_transaction(trans
, root
, ret
);
1200 key
.offset
= (u64
)-1;
1201 pending
->snap
= btrfs_read_fs_root_no_name(root
->fs_info
, &key
);
1202 if (IS_ERR(pending
->snap
)) {
1203 ret
= PTR_ERR(pending
->snap
);
1204 btrfs_abort_transaction(trans
, root
, ret
);
1208 ret
= btrfs_reloc_post_snapshot(trans
, pending
);
1210 btrfs_abort_transaction(trans
, root
, ret
);
1214 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1216 btrfs_abort_transaction(trans
, root
, ret
);
1220 ret
= btrfs_insert_dir_item(trans
, parent_root
,
1221 dentry
->d_name
.name
, dentry
->d_name
.len
,
1223 BTRFS_FT_DIR
, index
);
1224 /* We have check then name at the beginning, so it is impossible. */
1225 BUG_ON(ret
== -EEXIST
|| ret
== -EOVERFLOW
);
1227 btrfs_abort_transaction(trans
, root
, ret
);
1231 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
1232 dentry
->d_name
.len
* 2);
1233 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
1234 ret
= btrfs_update_inode_fallback(trans
, parent_root
, parent_inode
);
1236 btrfs_abort_transaction(trans
, root
, ret
);
1239 trans
->block_rsv
= rsv
;
1241 kfree(new_root_item
);
1242 root_item_alloc_fail
:
1243 btrfs_free_path(path
);
1245 btrfs_block_rsv_release(root
, &pending
->block_rsv
, (u64
)-1);
1250 * create all the snapshots we've scheduled for creation
1252 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
,
1253 struct btrfs_fs_info
*fs_info
)
1255 struct btrfs_pending_snapshot
*pending
;
1256 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
1258 list_for_each_entry(pending
, head
, list
)
1259 create_pending_snapshot(trans
, fs_info
, pending
);
1263 static void update_super_roots(struct btrfs_root
*root
)
1265 struct btrfs_root_item
*root_item
;
1266 struct btrfs_super_block
*super
;
1268 super
= root
->fs_info
->super_copy
;
1270 root_item
= &root
->fs_info
->chunk_root
->root_item
;
1271 super
->chunk_root
= root_item
->bytenr
;
1272 super
->chunk_root_generation
= root_item
->generation
;
1273 super
->chunk_root_level
= root_item
->level
;
1275 root_item
= &root
->fs_info
->tree_root
->root_item
;
1276 super
->root
= root_item
->bytenr
;
1277 super
->generation
= root_item
->generation
;
1278 super
->root_level
= root_item
->level
;
1279 if (btrfs_test_opt(root
, SPACE_CACHE
))
1280 super
->cache_generation
= root_item
->generation
;
1283 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
1286 spin_lock(&info
->trans_lock
);
1287 if (info
->running_transaction
)
1288 ret
= info
->running_transaction
->in_commit
;
1289 spin_unlock(&info
->trans_lock
);
1293 int btrfs_transaction_blocked(struct btrfs_fs_info
*info
)
1296 spin_lock(&info
->trans_lock
);
1297 if (info
->running_transaction
)
1298 ret
= info
->running_transaction
->blocked
;
1299 spin_unlock(&info
->trans_lock
);
1304 * wait for the current transaction commit to start and block subsequent
1307 static void wait_current_trans_commit_start(struct btrfs_root
*root
,
1308 struct btrfs_transaction
*trans
)
1310 wait_event(root
->fs_info
->transaction_blocked_wait
, trans
->in_commit
);
1314 * wait for the current transaction to start and then become unblocked.
1317 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root
*root
,
1318 struct btrfs_transaction
*trans
)
1320 wait_event(root
->fs_info
->transaction_wait
,
1321 trans
->commit_done
|| (trans
->in_commit
&& !trans
->blocked
));
1325 * commit transactions asynchronously. once btrfs_commit_transaction_async
1326 * returns, any subsequent transaction will not be allowed to join.
1328 struct btrfs_async_commit
{
1329 struct btrfs_trans_handle
*newtrans
;
1330 struct btrfs_root
*root
;
1331 struct work_struct work
;
1334 static void do_async_commit(struct work_struct
*work
)
1336 struct btrfs_async_commit
*ac
=
1337 container_of(work
, struct btrfs_async_commit
, work
);
1340 * We've got freeze protection passed with the transaction.
1341 * Tell lockdep about it.
1343 if (ac
->newtrans
->type
< TRANS_JOIN_NOLOCK
)
1345 &ac
->root
->fs_info
->sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
1348 current
->journal_info
= ac
->newtrans
;
1350 btrfs_commit_transaction(ac
->newtrans
, ac
->root
);
1354 int btrfs_commit_transaction_async(struct btrfs_trans_handle
*trans
,
1355 struct btrfs_root
*root
,
1356 int wait_for_unblock
)
1358 struct btrfs_async_commit
*ac
;
1359 struct btrfs_transaction
*cur_trans
;
1361 ac
= kmalloc(sizeof(*ac
), GFP_NOFS
);
1365 INIT_WORK(&ac
->work
, do_async_commit
);
1367 ac
->newtrans
= btrfs_join_transaction(root
);
1368 if (IS_ERR(ac
->newtrans
)) {
1369 int err
= PTR_ERR(ac
->newtrans
);
1374 /* take transaction reference */
1375 cur_trans
= trans
->transaction
;
1376 atomic_inc(&cur_trans
->use_count
);
1378 btrfs_end_transaction(trans
, root
);
1381 * Tell lockdep we've released the freeze rwsem, since the
1382 * async commit thread will be the one to unlock it.
1384 if (trans
->type
< TRANS_JOIN_NOLOCK
)
1386 &root
->fs_info
->sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
1389 schedule_work(&ac
->work
);
1391 /* wait for transaction to start and unblock */
1392 if (wait_for_unblock
)
1393 wait_current_trans_commit_start_and_unblock(root
, cur_trans
);
1395 wait_current_trans_commit_start(root
, cur_trans
);
1397 if (current
->journal_info
== trans
)
1398 current
->journal_info
= NULL
;
1400 put_transaction(cur_trans
);
1405 static void cleanup_transaction(struct btrfs_trans_handle
*trans
,
1406 struct btrfs_root
*root
, int err
)
1408 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1410 WARN_ON(trans
->use_count
> 1);
1412 btrfs_abort_transaction(trans
, root
, err
);
1414 spin_lock(&root
->fs_info
->trans_lock
);
1415 list_del_init(&cur_trans
->list
);
1416 if (cur_trans
== root
->fs_info
->running_transaction
) {
1417 root
->fs_info
->running_transaction
= NULL
;
1418 root
->fs_info
->trans_no_join
= 0;
1420 spin_unlock(&root
->fs_info
->trans_lock
);
1422 btrfs_cleanup_one_transaction(trans
->transaction
, root
);
1424 put_transaction(cur_trans
);
1425 put_transaction(cur_trans
);
1427 trace_btrfs_transaction_commit(root
);
1429 btrfs_scrub_continue(root
);
1431 if (current
->journal_info
== trans
)
1432 current
->journal_info
= NULL
;
1434 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1437 static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle
*trans
,
1438 struct btrfs_root
*root
)
1440 int flush_on_commit
= btrfs_test_opt(root
, FLUSHONCOMMIT
);
1441 int snap_pending
= 0;
1444 if (!flush_on_commit
) {
1445 spin_lock(&root
->fs_info
->trans_lock
);
1446 if (!list_empty(&trans
->transaction
->pending_snapshots
))
1448 spin_unlock(&root
->fs_info
->trans_lock
);
1451 if (flush_on_commit
|| snap_pending
) {
1452 ret
= btrfs_start_delalloc_inodes(root
, 1);
1455 btrfs_wait_ordered_extents(root
, 1);
1458 ret
= btrfs_run_delayed_items(trans
, root
);
1463 * running the delayed items may have added new refs. account
1464 * them now so that they hinder processing of more delayed refs
1465 * as little as possible.
1467 btrfs_delayed_refs_qgroup_accounting(trans
, root
->fs_info
);
1470 * rename don't use btrfs_join_transaction, so, once we
1471 * set the transaction to blocked above, we aren't going
1472 * to get any new ordered operations. We can safely run
1473 * it here and no for sure that nothing new will be added
1476 ret
= btrfs_run_ordered_operations(trans
, root
, 1);
1482 * btrfs_transaction state sequence:
1483 * in_commit = 0, blocked = 0 (initial)
1484 * in_commit = 1, blocked = 1
1488 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
,
1489 struct btrfs_root
*root
)
1491 unsigned long joined
= 0;
1492 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1493 struct btrfs_transaction
*prev_trans
= NULL
;
1496 int should_grow
= 0;
1497 unsigned long now
= get_seconds();
1499 ret
= btrfs_run_ordered_operations(trans
, root
, 0);
1501 btrfs_abort_transaction(trans
, root
, ret
);
1502 btrfs_end_transaction(trans
, root
);
1506 /* Stop the commit early if ->aborted is set */
1507 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1508 ret
= cur_trans
->aborted
;
1509 btrfs_end_transaction(trans
, root
);
1513 /* make a pass through all the delayed refs we have so far
1514 * any runnings procs may add more while we are here
1516 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1518 btrfs_end_transaction(trans
, root
);
1522 btrfs_trans_release_metadata(trans
, root
);
1523 trans
->block_rsv
= NULL
;
1525 cur_trans
= trans
->transaction
;
1528 * set the flushing flag so procs in this transaction have to
1529 * start sending their work down.
1531 cur_trans
->delayed_refs
.flushing
= 1;
1533 if (!list_empty(&trans
->new_bgs
))
1534 btrfs_create_pending_block_groups(trans
, root
);
1536 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1538 btrfs_end_transaction(trans
, root
);
1542 spin_lock(&cur_trans
->commit_lock
);
1543 if (cur_trans
->in_commit
) {
1544 spin_unlock(&cur_trans
->commit_lock
);
1545 atomic_inc(&cur_trans
->use_count
);
1546 ret
= btrfs_end_transaction(trans
, root
);
1548 wait_for_commit(root
, cur_trans
);
1550 put_transaction(cur_trans
);
1555 trans
->transaction
->in_commit
= 1;
1556 trans
->transaction
->blocked
= 1;
1557 spin_unlock(&cur_trans
->commit_lock
);
1558 wake_up(&root
->fs_info
->transaction_blocked_wait
);
1560 spin_lock(&root
->fs_info
->trans_lock
);
1561 if (cur_trans
->list
.prev
!= &root
->fs_info
->trans_list
) {
1562 prev_trans
= list_entry(cur_trans
->list
.prev
,
1563 struct btrfs_transaction
, list
);
1564 if (!prev_trans
->commit_done
) {
1565 atomic_inc(&prev_trans
->use_count
);
1566 spin_unlock(&root
->fs_info
->trans_lock
);
1568 wait_for_commit(root
, prev_trans
);
1570 put_transaction(prev_trans
);
1572 spin_unlock(&root
->fs_info
->trans_lock
);
1575 spin_unlock(&root
->fs_info
->trans_lock
);
1578 if (!btrfs_test_opt(root
, SSD
) &&
1579 (now
< cur_trans
->start_time
|| now
- cur_trans
->start_time
< 1))
1583 joined
= cur_trans
->num_joined
;
1585 WARN_ON(cur_trans
!= trans
->transaction
);
1587 ret
= btrfs_flush_all_pending_stuffs(trans
, root
);
1589 goto cleanup_transaction
;
1591 prepare_to_wait(&cur_trans
->writer_wait
, &wait
,
1592 TASK_UNINTERRUPTIBLE
);
1594 if (atomic_read(&cur_trans
->num_writers
) > 1)
1595 schedule_timeout(MAX_SCHEDULE_TIMEOUT
);
1596 else if (should_grow
)
1597 schedule_timeout(1);
1599 finish_wait(&cur_trans
->writer_wait
, &wait
);
1600 } while (atomic_read(&cur_trans
->num_writers
) > 1 ||
1601 (should_grow
&& cur_trans
->num_joined
!= joined
));
1603 ret
= btrfs_flush_all_pending_stuffs(trans
, root
);
1605 goto cleanup_transaction
;
1608 * Ok now we need to make sure to block out any other joins while we
1609 * commit the transaction. We could have started a join before setting
1610 * no_join so make sure to wait for num_writers to == 1 again.
1612 spin_lock(&root
->fs_info
->trans_lock
);
1613 root
->fs_info
->trans_no_join
= 1;
1614 spin_unlock(&root
->fs_info
->trans_lock
);
1615 wait_event(cur_trans
->writer_wait
,
1616 atomic_read(&cur_trans
->num_writers
) == 1);
1618 /* ->aborted might be set after the previous check, so check it */
1619 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1620 ret
= cur_trans
->aborted
;
1621 goto cleanup_transaction
;
1624 * the reloc mutex makes sure that we stop
1625 * the balancing code from coming in and moving
1626 * extents around in the middle of the commit
1628 mutex_lock(&root
->fs_info
->reloc_mutex
);
1631 * We needn't worry about the delayed items because we will
1632 * deal with them in create_pending_snapshot(), which is the
1633 * core function of the snapshot creation.
1635 ret
= create_pending_snapshots(trans
, root
->fs_info
);
1637 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1638 goto cleanup_transaction
;
1642 * We insert the dir indexes of the snapshots and update the inode
1643 * of the snapshots' parents after the snapshot creation, so there
1644 * are some delayed items which are not dealt with. Now deal with
1647 * We needn't worry that this operation will corrupt the snapshots,
1648 * because all the tree which are snapshoted will be forced to COW
1649 * the nodes and leaves.
1651 ret
= btrfs_run_delayed_items(trans
, root
);
1653 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1654 goto cleanup_transaction
;
1657 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1659 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1660 goto cleanup_transaction
;
1664 * make sure none of the code above managed to slip in a
1667 btrfs_assert_delayed_root_empty(root
);
1669 WARN_ON(cur_trans
!= trans
->transaction
);
1671 btrfs_scrub_pause(root
);
1672 /* btrfs_commit_tree_roots is responsible for getting the
1673 * various roots consistent with each other. Every pointer
1674 * in the tree of tree roots has to point to the most up to date
1675 * root for every subvolume and other tree. So, we have to keep
1676 * the tree logging code from jumping in and changing any
1679 * At this point in the commit, there can't be any tree-log
1680 * writers, but a little lower down we drop the trans mutex
1681 * and let new people in. By holding the tree_log_mutex
1682 * from now until after the super is written, we avoid races
1683 * with the tree-log code.
1685 mutex_lock(&root
->fs_info
->tree_log_mutex
);
1687 ret
= commit_fs_roots(trans
, root
);
1689 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1690 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1691 goto cleanup_transaction
;
1694 /* commit_fs_roots gets rid of all the tree log roots, it is now
1695 * safe to free the root of tree log roots
1697 btrfs_free_log_root_tree(trans
, root
->fs_info
);
1699 ret
= commit_cowonly_roots(trans
, root
);
1701 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1702 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1703 goto cleanup_transaction
;
1707 * The tasks which save the space cache and inode cache may also
1708 * update ->aborted, check it.
1710 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1711 ret
= cur_trans
->aborted
;
1712 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1713 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1714 goto cleanup_transaction
;
1717 btrfs_prepare_extent_commit(trans
, root
);
1719 cur_trans
= root
->fs_info
->running_transaction
;
1721 btrfs_set_root_node(&root
->fs_info
->tree_root
->root_item
,
1722 root
->fs_info
->tree_root
->node
);
1723 switch_commit_root(root
->fs_info
->tree_root
);
1725 btrfs_set_root_node(&root
->fs_info
->chunk_root
->root_item
,
1726 root
->fs_info
->chunk_root
->node
);
1727 switch_commit_root(root
->fs_info
->chunk_root
);
1729 assert_qgroups_uptodate(trans
);
1730 update_super_roots(root
);
1732 if (!root
->fs_info
->log_root_recovering
) {
1733 btrfs_set_super_log_root(root
->fs_info
->super_copy
, 0);
1734 btrfs_set_super_log_root_level(root
->fs_info
->super_copy
, 0);
1737 memcpy(root
->fs_info
->super_for_commit
, root
->fs_info
->super_copy
,
1738 sizeof(*root
->fs_info
->super_copy
));
1740 trans
->transaction
->blocked
= 0;
1741 spin_lock(&root
->fs_info
->trans_lock
);
1742 root
->fs_info
->running_transaction
= NULL
;
1743 root
->fs_info
->trans_no_join
= 0;
1744 spin_unlock(&root
->fs_info
->trans_lock
);
1745 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1747 wake_up(&root
->fs_info
->transaction_wait
);
1749 ret
= btrfs_write_and_wait_transaction(trans
, root
);
1751 btrfs_error(root
->fs_info
, ret
,
1752 "Error while writing out transaction.");
1753 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1754 goto cleanup_transaction
;
1757 ret
= write_ctree_super(trans
, root
, 0);
1759 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1760 goto cleanup_transaction
;
1764 * the super is written, we can safely allow the tree-loggers
1765 * to go about their business
1767 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1769 btrfs_finish_extent_commit(trans
, root
);
1771 cur_trans
->commit_done
= 1;
1773 root
->fs_info
->last_trans_committed
= cur_trans
->transid
;
1775 wake_up(&cur_trans
->commit_wait
);
1777 spin_lock(&root
->fs_info
->trans_lock
);
1778 list_del_init(&cur_trans
->list
);
1779 spin_unlock(&root
->fs_info
->trans_lock
);
1781 put_transaction(cur_trans
);
1782 put_transaction(cur_trans
);
1784 if (trans
->type
< TRANS_JOIN_NOLOCK
)
1785 sb_end_intwrite(root
->fs_info
->sb
);
1787 trace_btrfs_transaction_commit(root
);
1789 btrfs_scrub_continue(root
);
1791 if (current
->journal_info
== trans
)
1792 current
->journal_info
= NULL
;
1794 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1796 if (current
!= root
->fs_info
->transaction_kthread
)
1797 btrfs_run_delayed_iputs(root
);
1801 cleanup_transaction
:
1802 btrfs_trans_release_metadata(trans
, root
);
1803 trans
->block_rsv
= NULL
;
1804 btrfs_printk(root
->fs_info
, "Skipping commit of aborted transaction.\n");
1806 if (current
->journal_info
== trans
)
1807 current
->journal_info
= NULL
;
1808 cleanup_transaction(trans
, root
, ret
);
1814 * interface function to delete all the snapshots we have scheduled for deletion
1816 int btrfs_clean_old_snapshots(struct btrfs_root
*root
)
1819 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1821 spin_lock(&fs_info
->trans_lock
);
1822 list_splice_init(&fs_info
->dead_roots
, &list
);
1823 spin_unlock(&fs_info
->trans_lock
);
1825 while (!list_empty(&list
)) {
1828 root
= list_entry(list
.next
, struct btrfs_root
, root_list
);
1829 list_del(&root
->root_list
);
1831 btrfs_kill_all_delayed_nodes(root
);
1833 if (btrfs_header_backref_rev(root
->node
) <
1834 BTRFS_MIXED_BACKREF_REV
)
1835 ret
= btrfs_drop_snapshot(root
, NULL
, 0, 0);
1837 ret
=btrfs_drop_snapshot(root
, NULL
, 1, 0);