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
);
127 } else if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
128 spin_unlock(&fs_info
->trans_lock
);
129 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
133 atomic_set(&cur_trans
->num_writers
, 1);
134 cur_trans
->num_joined
= 0;
135 init_waitqueue_head(&cur_trans
->writer_wait
);
136 init_waitqueue_head(&cur_trans
->commit_wait
);
137 cur_trans
->in_commit
= 0;
138 cur_trans
->blocked
= 0;
140 * One for this trans handle, one so it will live on until we
141 * commit the transaction.
143 atomic_set(&cur_trans
->use_count
, 2);
144 cur_trans
->commit_done
= 0;
145 cur_trans
->start_time
= get_seconds();
147 cur_trans
->delayed_refs
.root
= RB_ROOT
;
148 cur_trans
->delayed_refs
.num_entries
= 0;
149 cur_trans
->delayed_refs
.num_heads_ready
= 0;
150 cur_trans
->delayed_refs
.num_heads
= 0;
151 cur_trans
->delayed_refs
.flushing
= 0;
152 cur_trans
->delayed_refs
.run_delayed_start
= 0;
155 * although the tree mod log is per file system and not per transaction,
156 * the log must never go across transaction boundaries.
159 if (!list_empty(&fs_info
->tree_mod_seq_list
))
160 WARN(1, KERN_ERR
"btrfs: tree_mod_seq_list not empty when "
161 "creating a fresh transaction\n");
162 if (!RB_EMPTY_ROOT(&fs_info
->tree_mod_log
))
163 WARN(1, KERN_ERR
"btrfs: tree_mod_log rb tree not empty when "
164 "creating a fresh transaction\n");
165 atomic_set(&fs_info
->tree_mod_seq
, 0);
167 spin_lock_init(&cur_trans
->commit_lock
);
168 spin_lock_init(&cur_trans
->delayed_refs
.lock
);
169 atomic_set(&cur_trans
->delayed_refs
.procs_running_refs
, 0);
170 atomic_set(&cur_trans
->delayed_refs
.ref_seq
, 0);
171 init_waitqueue_head(&cur_trans
->delayed_refs
.wait
);
173 INIT_LIST_HEAD(&cur_trans
->pending_snapshots
);
174 INIT_LIST_HEAD(&cur_trans
->ordered_operations
);
175 list_add_tail(&cur_trans
->list
, &fs_info
->trans_list
);
176 extent_io_tree_init(&cur_trans
->dirty_pages
,
177 fs_info
->btree_inode
->i_mapping
);
178 fs_info
->generation
++;
179 cur_trans
->transid
= fs_info
->generation
;
180 fs_info
->running_transaction
= cur_trans
;
181 cur_trans
->aborted
= 0;
182 spin_unlock(&fs_info
->trans_lock
);
188 * this does all the record keeping required to make sure that a reference
189 * counted root is properly recorded in a given transaction. This is required
190 * to make sure the old root from before we joined the transaction is deleted
191 * when the transaction commits
193 static int record_root_in_trans(struct btrfs_trans_handle
*trans
,
194 struct btrfs_root
*root
)
196 if (root
->ref_cows
&& root
->last_trans
< trans
->transid
) {
197 WARN_ON(root
== root
->fs_info
->extent_root
);
198 WARN_ON(root
->commit_root
!= root
->node
);
201 * see below for in_trans_setup usage rules
202 * we have the reloc mutex held now, so there
203 * is only one writer in this function
205 root
->in_trans_setup
= 1;
207 /* make sure readers find in_trans_setup before
208 * they find our root->last_trans update
212 spin_lock(&root
->fs_info
->fs_roots_radix_lock
);
213 if (root
->last_trans
== trans
->transid
) {
214 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
217 radix_tree_tag_set(&root
->fs_info
->fs_roots_radix
,
218 (unsigned long)root
->root_key
.objectid
,
219 BTRFS_ROOT_TRANS_TAG
);
220 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
221 root
->last_trans
= trans
->transid
;
223 /* this is pretty tricky. We don't want to
224 * take the relocation lock in btrfs_record_root_in_trans
225 * unless we're really doing the first setup for this root in
228 * Normally we'd use root->last_trans as a flag to decide
229 * if we want to take the expensive mutex.
231 * But, we have to set root->last_trans before we
232 * init the relocation root, otherwise, we trip over warnings
233 * in ctree.c. The solution used here is to flag ourselves
234 * with root->in_trans_setup. When this is 1, we're still
235 * fixing up the reloc trees and everyone must wait.
237 * When this is zero, they can trust root->last_trans and fly
238 * through btrfs_record_root_in_trans without having to take the
239 * lock. smp_wmb() makes sure that all the writes above are
240 * done before we pop in the zero below
242 btrfs_init_reloc_root(trans
, root
);
244 root
->in_trans_setup
= 0;
250 int btrfs_record_root_in_trans(struct btrfs_trans_handle
*trans
,
251 struct btrfs_root
*root
)
257 * see record_root_in_trans for comments about in_trans_setup usage
261 if (root
->last_trans
== trans
->transid
&&
262 !root
->in_trans_setup
)
265 mutex_lock(&root
->fs_info
->reloc_mutex
);
266 record_root_in_trans(trans
, root
);
267 mutex_unlock(&root
->fs_info
->reloc_mutex
);
272 /* wait for commit against the current transaction to become unblocked
273 * when this is done, it is safe to start a new transaction, but the current
274 * transaction might not be fully on disk.
276 static void wait_current_trans(struct btrfs_root
*root
)
278 struct btrfs_transaction
*cur_trans
;
280 spin_lock(&root
->fs_info
->trans_lock
);
281 cur_trans
= root
->fs_info
->running_transaction
;
282 if (cur_trans
&& cur_trans
->blocked
) {
283 atomic_inc(&cur_trans
->use_count
);
284 spin_unlock(&root
->fs_info
->trans_lock
);
286 wait_event(root
->fs_info
->transaction_wait
,
287 !cur_trans
->blocked
);
288 put_transaction(cur_trans
);
290 spin_unlock(&root
->fs_info
->trans_lock
);
294 static int may_wait_transaction(struct btrfs_root
*root
, int type
)
296 if (root
->fs_info
->log_root_recovering
)
299 if (type
== TRANS_USERSPACE
)
302 if (type
== TRANS_START
&&
303 !atomic_read(&root
->fs_info
->open_ioctl_trans
))
309 static struct btrfs_trans_handle
*
310 start_transaction(struct btrfs_root
*root
, u64 num_items
, int type
,
311 enum btrfs_reserve_flush_enum flush
)
313 struct btrfs_trans_handle
*h
;
314 struct btrfs_transaction
*cur_trans
;
317 u64 qgroup_reserved
= 0;
319 if (test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
320 return ERR_PTR(-EROFS
);
322 if (current
->journal_info
) {
323 WARN_ON(type
!= TRANS_JOIN
&& type
!= TRANS_JOIN_NOLOCK
);
324 h
= current
->journal_info
;
326 WARN_ON(h
->use_count
> 2);
327 h
->orig_rsv
= h
->block_rsv
;
333 * Do the reservation before we join the transaction so we can do all
334 * the appropriate flushing if need be.
336 if (num_items
> 0 && root
!= root
->fs_info
->chunk_root
) {
337 if (root
->fs_info
->quota_enabled
&&
338 is_fstree(root
->root_key
.objectid
)) {
339 qgroup_reserved
= num_items
* root
->leafsize
;
340 ret
= btrfs_qgroup_reserve(root
, qgroup_reserved
);
345 num_bytes
= btrfs_calc_trans_metadata_size(root
, num_items
);
346 ret
= btrfs_block_rsv_add(root
,
347 &root
->fs_info
->trans_block_rsv
,
353 h
= kmem_cache_alloc(btrfs_trans_handle_cachep
, GFP_NOFS
);
360 * If we are JOIN_NOLOCK we're already committing a transaction and
361 * waiting on this guy, so we don't need to do the sb_start_intwrite
362 * because we're already holding a ref. We need this because we could
363 * have raced in and did an fsync() on a file which can kick a commit
364 * and then we deadlock with somebody doing a freeze.
366 * If we are ATTACH, it means we just want to catch the current
367 * transaction and commit it, so we needn't do sb_start_intwrite().
369 if (type
< TRANS_JOIN_NOLOCK
)
370 sb_start_intwrite(root
->fs_info
->sb
);
372 if (may_wait_transaction(root
, type
))
373 wait_current_trans(root
);
376 ret
= join_transaction(root
, type
);
378 wait_current_trans(root
);
379 if (unlikely(type
== TRANS_ATTACH
))
382 } while (ret
== -EBUSY
);
385 /* We must get the transaction if we are JOIN_NOLOCK. */
386 BUG_ON(type
== TRANS_JOIN_NOLOCK
);
390 cur_trans
= root
->fs_info
->running_transaction
;
392 h
->transid
= cur_trans
->transid
;
393 h
->transaction
= cur_trans
;
395 h
->bytes_reserved
= 0;
397 h
->delayed_ref_updates
= 0;
403 h
->qgroup_reserved
= 0;
404 h
->delayed_ref_elem
.seq
= 0;
406 h
->allocating_chunk
= false;
407 INIT_LIST_HEAD(&h
->qgroup_ref_list
);
408 INIT_LIST_HEAD(&h
->new_bgs
);
411 if (cur_trans
->blocked
&& may_wait_transaction(root
, type
)) {
412 btrfs_commit_transaction(h
, root
);
417 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
418 h
->transid
, num_bytes
, 1);
419 h
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
420 h
->bytes_reserved
= num_bytes
;
422 h
->qgroup_reserved
= qgroup_reserved
;
425 btrfs_record_root_in_trans(h
, root
);
427 if (!current
->journal_info
&& type
!= TRANS_USERSPACE
)
428 current
->journal_info
= h
;
432 if (type
< TRANS_JOIN_NOLOCK
)
433 sb_end_intwrite(root
->fs_info
->sb
);
434 kmem_cache_free(btrfs_trans_handle_cachep
, h
);
437 btrfs_block_rsv_release(root
, &root
->fs_info
->trans_block_rsv
,
441 btrfs_qgroup_free(root
, qgroup_reserved
);
445 struct btrfs_trans_handle
*btrfs_start_transaction(struct btrfs_root
*root
,
448 return start_transaction(root
, num_items
, TRANS_START
,
449 BTRFS_RESERVE_FLUSH_ALL
);
452 struct btrfs_trans_handle
*btrfs_start_transaction_lflush(
453 struct btrfs_root
*root
, int num_items
)
455 return start_transaction(root
, num_items
, TRANS_START
,
456 BTRFS_RESERVE_FLUSH_LIMIT
);
459 struct btrfs_trans_handle
*btrfs_join_transaction(struct btrfs_root
*root
)
461 return start_transaction(root
, 0, TRANS_JOIN
, 0);
464 struct btrfs_trans_handle
*btrfs_join_transaction_nolock(struct btrfs_root
*root
)
466 return start_transaction(root
, 0, TRANS_JOIN_NOLOCK
, 0);
469 struct btrfs_trans_handle
*btrfs_start_ioctl_transaction(struct btrfs_root
*root
)
471 return start_transaction(root
, 0, TRANS_USERSPACE
, 0);
475 * btrfs_attach_transaction() - catch the running transaction
477 * It is used when we want to commit the current the transaction, but
478 * don't want to start a new one.
480 * Note: If this function return -ENOENT, it just means there is no
481 * running transaction. But it is possible that the inactive transaction
482 * is still in the memory, not fully on disk. If you hope there is no
483 * inactive transaction in the fs when -ENOENT is returned, you should
485 * btrfs_attach_transaction_barrier()
487 struct btrfs_trans_handle
*btrfs_attach_transaction(struct btrfs_root
*root
)
489 return start_transaction(root
, 0, TRANS_ATTACH
, 0);
493 * btrfs_attach_transaction() - catch the running transaction
495 * It is similar to the above function, the differentia is this one
496 * will wait for all the inactive transactions until they fully
499 struct btrfs_trans_handle
*
500 btrfs_attach_transaction_barrier(struct btrfs_root
*root
)
502 struct btrfs_trans_handle
*trans
;
504 trans
= start_transaction(root
, 0, TRANS_ATTACH
, 0);
505 if (IS_ERR(trans
) && PTR_ERR(trans
) == -ENOENT
)
506 btrfs_wait_for_commit(root
, 0);
511 /* wait for a transaction commit to be fully complete */
512 static noinline
void wait_for_commit(struct btrfs_root
*root
,
513 struct btrfs_transaction
*commit
)
515 wait_event(commit
->commit_wait
, commit
->commit_done
);
518 int btrfs_wait_for_commit(struct btrfs_root
*root
, u64 transid
)
520 struct btrfs_transaction
*cur_trans
= NULL
, *t
;
524 if (transid
<= root
->fs_info
->last_trans_committed
)
528 /* find specified transaction */
529 spin_lock(&root
->fs_info
->trans_lock
);
530 list_for_each_entry(t
, &root
->fs_info
->trans_list
, list
) {
531 if (t
->transid
== transid
) {
533 atomic_inc(&cur_trans
->use_count
);
537 if (t
->transid
> transid
) {
542 spin_unlock(&root
->fs_info
->trans_lock
);
543 /* The specified transaction doesn't exist */
547 /* find newest transaction that is committing | committed */
548 spin_lock(&root
->fs_info
->trans_lock
);
549 list_for_each_entry_reverse(t
, &root
->fs_info
->trans_list
,
555 atomic_inc(&cur_trans
->use_count
);
559 spin_unlock(&root
->fs_info
->trans_lock
);
561 goto out
; /* nothing committing|committed */
564 wait_for_commit(root
, cur_trans
);
565 put_transaction(cur_trans
);
570 void btrfs_throttle(struct btrfs_root
*root
)
572 if (!atomic_read(&root
->fs_info
->open_ioctl_trans
))
573 wait_current_trans(root
);
576 static int should_end_transaction(struct btrfs_trans_handle
*trans
,
577 struct btrfs_root
*root
)
581 ret
= btrfs_block_rsv_check(root
, &root
->fs_info
->global_block_rsv
, 5);
585 int btrfs_should_end_transaction(struct btrfs_trans_handle
*trans
,
586 struct btrfs_root
*root
)
588 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
593 if (cur_trans
->blocked
|| cur_trans
->delayed_refs
.flushing
)
596 updates
= trans
->delayed_ref_updates
;
597 trans
->delayed_ref_updates
= 0;
599 err
= btrfs_run_delayed_refs(trans
, root
, updates
);
600 if (err
) /* Error code will also eval true */
604 return should_end_transaction(trans
, root
);
607 static int __btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
608 struct btrfs_root
*root
, int throttle
)
610 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
611 struct btrfs_fs_info
*info
= root
->fs_info
;
613 int lock
= (trans
->type
!= TRANS_JOIN_NOLOCK
);
616 if (--trans
->use_count
) {
617 trans
->block_rsv
= trans
->orig_rsv
;
622 * do the qgroup accounting as early as possible
624 err
= btrfs_delayed_refs_qgroup_accounting(trans
, info
);
626 btrfs_trans_release_metadata(trans
, root
);
627 trans
->block_rsv
= NULL
;
629 if (trans
->qgroup_reserved
) {
631 * the same root has to be passed here between start_transaction
632 * and end_transaction. Subvolume quota depends on this.
634 btrfs_qgroup_free(trans
->root
, trans
->qgroup_reserved
);
635 trans
->qgroup_reserved
= 0;
638 if (!list_empty(&trans
->new_bgs
))
639 btrfs_create_pending_block_groups(trans
, root
);
642 unsigned long cur
= trans
->delayed_ref_updates
;
643 trans
->delayed_ref_updates
= 0;
645 trans
->transaction
->delayed_refs
.num_heads_ready
> 64) {
646 trans
->delayed_ref_updates
= 0;
647 btrfs_run_delayed_refs(trans
, root
, cur
);
654 btrfs_trans_release_metadata(trans
, root
);
655 trans
->block_rsv
= NULL
;
657 if (!list_empty(&trans
->new_bgs
))
658 btrfs_create_pending_block_groups(trans
, root
);
660 if (lock
&& !atomic_read(&root
->fs_info
->open_ioctl_trans
) &&
661 should_end_transaction(trans
, root
)) {
662 trans
->transaction
->blocked
= 1;
666 if (lock
&& cur_trans
->blocked
&& !cur_trans
->in_commit
) {
669 * We may race with somebody else here so end up having
670 * to call end_transaction on ourselves again, so inc
674 return btrfs_commit_transaction(trans
, root
);
676 wake_up_process(info
->transaction_kthread
);
680 if (trans
->type
< TRANS_JOIN_NOLOCK
)
681 sb_end_intwrite(root
->fs_info
->sb
);
683 WARN_ON(cur_trans
!= info
->running_transaction
);
684 WARN_ON(atomic_read(&cur_trans
->num_writers
) < 1);
685 atomic_dec(&cur_trans
->num_writers
);
688 if (waitqueue_active(&cur_trans
->writer_wait
))
689 wake_up(&cur_trans
->writer_wait
);
690 put_transaction(cur_trans
);
692 if (current
->journal_info
== trans
)
693 current
->journal_info
= NULL
;
696 btrfs_run_delayed_iputs(root
);
698 if (trans
->aborted
||
699 test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
701 assert_qgroups_uptodate(trans
);
703 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
707 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
708 struct btrfs_root
*root
)
712 ret
= __btrfs_end_transaction(trans
, root
, 0);
718 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
,
719 struct btrfs_root
*root
)
723 ret
= __btrfs_end_transaction(trans
, root
, 1);
729 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle
*trans
,
730 struct btrfs_root
*root
)
732 return __btrfs_end_transaction(trans
, root
, 1);
736 * when btree blocks are allocated, they have some corresponding bits set for
737 * them in one of two extent_io trees. This is used to make sure all of
738 * those extents are sent to disk but does not wait on them
740 int btrfs_write_marked_extents(struct btrfs_root
*root
,
741 struct extent_io_tree
*dirty_pages
, int mark
)
745 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
746 struct extent_state
*cached_state
= NULL
;
749 struct blk_plug plug
;
751 blk_start_plug(&plug
);
752 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
753 mark
, &cached_state
)) {
754 convert_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
,
755 mark
, &cached_state
, GFP_NOFS
);
757 err
= filemap_fdatawrite_range(mapping
, start
, end
);
765 blk_finish_plug(&plug
);
770 * when btree blocks are allocated, they have some corresponding bits set for
771 * them in one of two extent_io trees. This is used to make sure all of
772 * those extents are on disk for transaction or log commit. We wait
773 * on all the pages and clear them from the dirty pages state tree
775 int btrfs_wait_marked_extents(struct btrfs_root
*root
,
776 struct extent_io_tree
*dirty_pages
, int mark
)
780 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
781 struct extent_state
*cached_state
= NULL
;
785 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
786 EXTENT_NEED_WAIT
, &cached_state
)) {
787 clear_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
,
788 0, 0, &cached_state
, GFP_NOFS
);
789 err
= filemap_fdatawait_range(mapping
, start
, end
);
801 * when btree blocks are allocated, they have some corresponding bits set for
802 * them in one of two extent_io trees. This is used to make sure all of
803 * those extents are on disk for transaction or log commit
805 int btrfs_write_and_wait_marked_extents(struct btrfs_root
*root
,
806 struct extent_io_tree
*dirty_pages
, int mark
)
811 ret
= btrfs_write_marked_extents(root
, dirty_pages
, mark
);
812 ret2
= btrfs_wait_marked_extents(root
, dirty_pages
, mark
);
821 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
,
822 struct btrfs_root
*root
)
824 if (!trans
|| !trans
->transaction
) {
825 struct inode
*btree_inode
;
826 btree_inode
= root
->fs_info
->btree_inode
;
827 return filemap_write_and_wait(btree_inode
->i_mapping
);
829 return btrfs_write_and_wait_marked_extents(root
,
830 &trans
->transaction
->dirty_pages
,
835 * this is used to update the root pointer in the tree of tree roots.
837 * But, in the case of the extent allocation tree, updating the root
838 * pointer may allocate blocks which may change the root of the extent
841 * So, this loops and repeats and makes sure the cowonly root didn't
842 * change while the root pointer was being updated in the metadata.
844 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
845 struct btrfs_root
*root
)
850 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
852 old_root_used
= btrfs_root_used(&root
->root_item
);
853 btrfs_write_dirty_block_groups(trans
, root
);
856 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
857 if (old_root_bytenr
== root
->node
->start
&&
858 old_root_used
== btrfs_root_used(&root
->root_item
))
861 btrfs_set_root_node(&root
->root_item
, root
->node
);
862 ret
= btrfs_update_root(trans
, tree_root
,
868 old_root_used
= btrfs_root_used(&root
->root_item
);
869 ret
= btrfs_write_dirty_block_groups(trans
, root
);
874 if (root
!= root
->fs_info
->extent_root
)
875 switch_commit_root(root
);
881 * update all the cowonly tree roots on disk
883 * The error handling in this function may not be obvious. Any of the
884 * failures will cause the file system to go offline. We still need
885 * to clean up the delayed refs.
887 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
,
888 struct btrfs_root
*root
)
890 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
891 struct list_head
*next
;
892 struct extent_buffer
*eb
;
895 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
899 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
900 ret
= btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
,
902 btrfs_tree_unlock(eb
);
903 free_extent_buffer(eb
);
908 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
912 ret
= btrfs_run_dev_stats(trans
, root
->fs_info
);
914 ret
= btrfs_run_dev_replace(trans
, root
->fs_info
);
917 ret
= btrfs_run_qgroups(trans
, root
->fs_info
);
920 /* run_qgroups might have added some more refs */
921 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
924 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
925 next
= fs_info
->dirty_cowonly_roots
.next
;
927 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
929 ret
= update_cowonly_root(trans
, root
);
934 down_write(&fs_info
->extent_commit_sem
);
935 switch_commit_root(fs_info
->extent_root
);
936 up_write(&fs_info
->extent_commit_sem
);
938 btrfs_after_dev_replace_commit(fs_info
);
944 * dead roots are old snapshots that need to be deleted. This allocates
945 * a dirty root struct and adds it into the list of dead roots that need to
948 int btrfs_add_dead_root(struct btrfs_root
*root
)
950 spin_lock(&root
->fs_info
->trans_lock
);
951 list_add(&root
->root_list
, &root
->fs_info
->dead_roots
);
952 spin_unlock(&root
->fs_info
->trans_lock
);
957 * update all the cowonly tree roots on disk
959 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
,
960 struct btrfs_root
*root
)
962 struct btrfs_root
*gang
[8];
963 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
968 spin_lock(&fs_info
->fs_roots_radix_lock
);
970 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
973 BTRFS_ROOT_TRANS_TAG
);
976 for (i
= 0; i
< ret
; i
++) {
978 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
979 (unsigned long)root
->root_key
.objectid
,
980 BTRFS_ROOT_TRANS_TAG
);
981 spin_unlock(&fs_info
->fs_roots_radix_lock
);
983 btrfs_free_log(trans
, root
);
984 btrfs_update_reloc_root(trans
, root
);
985 btrfs_orphan_commit_root(trans
, root
);
987 btrfs_save_ino_cache(root
, trans
);
989 /* see comments in should_cow_block() */
993 if (root
->commit_root
!= root
->node
) {
994 mutex_lock(&root
->fs_commit_mutex
);
995 switch_commit_root(root
);
996 btrfs_unpin_free_ino(root
);
997 mutex_unlock(&root
->fs_commit_mutex
);
999 btrfs_set_root_node(&root
->root_item
,
1003 err
= btrfs_update_root(trans
, fs_info
->tree_root
,
1006 spin_lock(&fs_info
->fs_roots_radix_lock
);
1011 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1016 * defrag a given btree.
1017 * Every leaf in the btree is read and defragged.
1019 int btrfs_defrag_root(struct btrfs_root
*root
)
1021 struct btrfs_fs_info
*info
= root
->fs_info
;
1022 struct btrfs_trans_handle
*trans
;
1025 if (xchg(&root
->defrag_running
, 1))
1029 trans
= btrfs_start_transaction(root
, 0);
1031 return PTR_ERR(trans
);
1033 ret
= btrfs_defrag_leaves(trans
, root
);
1035 btrfs_end_transaction(trans
, root
);
1036 btrfs_btree_balance_dirty(info
->tree_root
);
1039 if (btrfs_fs_closing(root
->fs_info
) || ret
!= -EAGAIN
)
1042 if (btrfs_defrag_cancelled(root
->fs_info
)) {
1043 printk(KERN_DEBUG
"btrfs: defrag_root cancelled\n");
1048 root
->defrag_running
= 0;
1053 * new snapshots need to be created at a very specific time in the
1054 * transaction commit. This does the actual creation.
1057 * If the error which may affect the commitment of the current transaction
1058 * happens, we should return the error number. If the error which just affect
1059 * the creation of the pending snapshots, just return 0.
1061 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
1062 struct btrfs_fs_info
*fs_info
,
1063 struct btrfs_pending_snapshot
*pending
)
1065 struct btrfs_key key
;
1066 struct btrfs_root_item
*new_root_item
;
1067 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1068 struct btrfs_root
*root
= pending
->root
;
1069 struct btrfs_root
*parent_root
;
1070 struct btrfs_block_rsv
*rsv
;
1071 struct inode
*parent_inode
;
1072 struct btrfs_path
*path
;
1073 struct btrfs_dir_item
*dir_item
;
1074 struct dentry
*dentry
;
1075 struct extent_buffer
*tmp
;
1076 struct extent_buffer
*old
;
1077 struct timespec cur_time
= CURRENT_TIME
;
1085 path
= btrfs_alloc_path();
1087 pending
->error
= -ENOMEM
;
1091 new_root_item
= kmalloc(sizeof(*new_root_item
), GFP_NOFS
);
1092 if (!new_root_item
) {
1093 pending
->error
= -ENOMEM
;
1094 goto root_item_alloc_fail
;
1097 pending
->error
= btrfs_find_free_objectid(tree_root
, &objectid
);
1099 goto no_free_objectid
;
1101 btrfs_reloc_pre_snapshot(trans
, pending
, &to_reserve
);
1103 if (to_reserve
> 0) {
1104 pending
->error
= btrfs_block_rsv_add(root
,
1105 &pending
->block_rsv
,
1107 BTRFS_RESERVE_NO_FLUSH
);
1109 goto no_free_objectid
;
1112 pending
->error
= btrfs_qgroup_inherit(trans
, fs_info
,
1113 root
->root_key
.objectid
,
1114 objectid
, pending
->inherit
);
1116 goto no_free_objectid
;
1118 key
.objectid
= objectid
;
1119 key
.offset
= (u64
)-1;
1120 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1122 rsv
= trans
->block_rsv
;
1123 trans
->block_rsv
= &pending
->block_rsv
;
1124 trans
->bytes_reserved
= trans
->block_rsv
->reserved
;
1126 dentry
= pending
->dentry
;
1127 parent_inode
= pending
->dir
;
1128 parent_root
= BTRFS_I(parent_inode
)->root
;
1129 record_root_in_trans(trans
, parent_root
);
1132 * insert the directory item
1134 ret
= btrfs_set_inode_index(parent_inode
, &index
);
1135 BUG_ON(ret
); /* -ENOMEM */
1137 /* check if there is a file/dir which has the same name. */
1138 dir_item
= btrfs_lookup_dir_item(NULL
, parent_root
, path
,
1139 btrfs_ino(parent_inode
),
1140 dentry
->d_name
.name
,
1141 dentry
->d_name
.len
, 0);
1142 if (dir_item
!= NULL
&& !IS_ERR(dir_item
)) {
1143 pending
->error
= -EEXIST
;
1144 goto dir_item_existed
;
1145 } else if (IS_ERR(dir_item
)) {
1146 ret
= PTR_ERR(dir_item
);
1147 btrfs_abort_transaction(trans
, root
, ret
);
1150 btrfs_release_path(path
);
1153 * pull in the delayed directory update
1154 * and the delayed inode item
1155 * otherwise we corrupt the FS during
1158 ret
= btrfs_run_delayed_items(trans
, root
);
1159 if (ret
) { /* Transaction aborted */
1160 btrfs_abort_transaction(trans
, root
, ret
);
1164 record_root_in_trans(trans
, root
);
1165 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
1166 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
1167 btrfs_check_and_init_root_item(new_root_item
);
1169 root_flags
= btrfs_root_flags(new_root_item
);
1170 if (pending
->readonly
)
1171 root_flags
|= BTRFS_ROOT_SUBVOL_RDONLY
;
1173 root_flags
&= ~BTRFS_ROOT_SUBVOL_RDONLY
;
1174 btrfs_set_root_flags(new_root_item
, root_flags
);
1176 btrfs_set_root_generation_v2(new_root_item
,
1178 uuid_le_gen(&new_uuid
);
1179 memcpy(new_root_item
->uuid
, new_uuid
.b
, BTRFS_UUID_SIZE
);
1180 memcpy(new_root_item
->parent_uuid
, root
->root_item
.uuid
,
1182 new_root_item
->otime
.sec
= cpu_to_le64(cur_time
.tv_sec
);
1183 new_root_item
->otime
.nsec
= cpu_to_le32(cur_time
.tv_nsec
);
1184 btrfs_set_root_otransid(new_root_item
, trans
->transid
);
1185 memset(&new_root_item
->stime
, 0, sizeof(new_root_item
->stime
));
1186 memset(&new_root_item
->rtime
, 0, sizeof(new_root_item
->rtime
));
1187 btrfs_set_root_stransid(new_root_item
, 0);
1188 btrfs_set_root_rtransid(new_root_item
, 0);
1190 old
= btrfs_lock_root_node(root
);
1191 ret
= btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
);
1193 btrfs_tree_unlock(old
);
1194 free_extent_buffer(old
);
1195 btrfs_abort_transaction(trans
, root
, ret
);
1199 btrfs_set_lock_blocking(old
);
1201 ret
= btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
1202 /* clean up in any case */
1203 btrfs_tree_unlock(old
);
1204 free_extent_buffer(old
);
1206 btrfs_abort_transaction(trans
, root
, ret
);
1210 /* see comments in should_cow_block() */
1211 root
->force_cow
= 1;
1214 btrfs_set_root_node(new_root_item
, tmp
);
1215 /* record when the snapshot was created in key.offset */
1216 key
.offset
= trans
->transid
;
1217 ret
= btrfs_insert_root(trans
, tree_root
, &key
, new_root_item
);
1218 btrfs_tree_unlock(tmp
);
1219 free_extent_buffer(tmp
);
1221 btrfs_abort_transaction(trans
, root
, ret
);
1226 * insert root back/forward references
1228 ret
= btrfs_add_root_ref(trans
, tree_root
, objectid
,
1229 parent_root
->root_key
.objectid
,
1230 btrfs_ino(parent_inode
), index
,
1231 dentry
->d_name
.name
, dentry
->d_name
.len
);
1233 btrfs_abort_transaction(trans
, root
, ret
);
1237 key
.offset
= (u64
)-1;
1238 pending
->snap
= btrfs_read_fs_root_no_name(root
->fs_info
, &key
);
1239 if (IS_ERR(pending
->snap
)) {
1240 ret
= PTR_ERR(pending
->snap
);
1241 btrfs_abort_transaction(trans
, root
, ret
);
1245 ret
= btrfs_reloc_post_snapshot(trans
, pending
);
1247 btrfs_abort_transaction(trans
, root
, ret
);
1251 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1253 btrfs_abort_transaction(trans
, root
, ret
);
1257 ret
= btrfs_insert_dir_item(trans
, parent_root
,
1258 dentry
->d_name
.name
, dentry
->d_name
.len
,
1260 BTRFS_FT_DIR
, index
);
1261 /* We have check then name at the beginning, so it is impossible. */
1262 BUG_ON(ret
== -EEXIST
|| ret
== -EOVERFLOW
);
1264 btrfs_abort_transaction(trans
, root
, ret
);
1268 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
1269 dentry
->d_name
.len
* 2);
1270 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
1271 ret
= btrfs_update_inode_fallback(trans
, parent_root
, parent_inode
);
1273 btrfs_abort_transaction(trans
, root
, ret
);
1275 pending
->error
= ret
;
1277 trans
->block_rsv
= rsv
;
1278 trans
->bytes_reserved
= 0;
1280 kfree(new_root_item
);
1281 root_item_alloc_fail
:
1282 btrfs_free_path(path
);
1287 * create all the snapshots we've scheduled for creation
1289 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
,
1290 struct btrfs_fs_info
*fs_info
)
1292 struct btrfs_pending_snapshot
*pending
, *next
;
1293 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
1296 list_for_each_entry_safe(pending
, next
, head
, list
) {
1297 list_del(&pending
->list
);
1298 ret
= create_pending_snapshot(trans
, fs_info
, pending
);
1305 static void update_super_roots(struct btrfs_root
*root
)
1307 struct btrfs_root_item
*root_item
;
1308 struct btrfs_super_block
*super
;
1310 super
= root
->fs_info
->super_copy
;
1312 root_item
= &root
->fs_info
->chunk_root
->root_item
;
1313 super
->chunk_root
= root_item
->bytenr
;
1314 super
->chunk_root_generation
= root_item
->generation
;
1315 super
->chunk_root_level
= root_item
->level
;
1317 root_item
= &root
->fs_info
->tree_root
->root_item
;
1318 super
->root
= root_item
->bytenr
;
1319 super
->generation
= root_item
->generation
;
1320 super
->root_level
= root_item
->level
;
1321 if (btrfs_test_opt(root
, SPACE_CACHE
))
1322 super
->cache_generation
= root_item
->generation
;
1325 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
1328 spin_lock(&info
->trans_lock
);
1329 if (info
->running_transaction
)
1330 ret
= info
->running_transaction
->in_commit
;
1331 spin_unlock(&info
->trans_lock
);
1335 int btrfs_transaction_blocked(struct btrfs_fs_info
*info
)
1338 spin_lock(&info
->trans_lock
);
1339 if (info
->running_transaction
)
1340 ret
= info
->running_transaction
->blocked
;
1341 spin_unlock(&info
->trans_lock
);
1346 * wait for the current transaction commit to start and block subsequent
1349 static void wait_current_trans_commit_start(struct btrfs_root
*root
,
1350 struct btrfs_transaction
*trans
)
1352 wait_event(root
->fs_info
->transaction_blocked_wait
, trans
->in_commit
);
1356 * wait for the current transaction to start and then become unblocked.
1359 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root
*root
,
1360 struct btrfs_transaction
*trans
)
1362 wait_event(root
->fs_info
->transaction_wait
,
1363 trans
->commit_done
|| (trans
->in_commit
&& !trans
->blocked
));
1367 * commit transactions asynchronously. once btrfs_commit_transaction_async
1368 * returns, any subsequent transaction will not be allowed to join.
1370 struct btrfs_async_commit
{
1371 struct btrfs_trans_handle
*newtrans
;
1372 struct btrfs_root
*root
;
1373 struct work_struct work
;
1376 static void do_async_commit(struct work_struct
*work
)
1378 struct btrfs_async_commit
*ac
=
1379 container_of(work
, struct btrfs_async_commit
, work
);
1382 * We've got freeze protection passed with the transaction.
1383 * Tell lockdep about it.
1385 if (ac
->newtrans
->type
< TRANS_JOIN_NOLOCK
)
1387 &ac
->root
->fs_info
->sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
1390 current
->journal_info
= ac
->newtrans
;
1392 btrfs_commit_transaction(ac
->newtrans
, ac
->root
);
1396 int btrfs_commit_transaction_async(struct btrfs_trans_handle
*trans
,
1397 struct btrfs_root
*root
,
1398 int wait_for_unblock
)
1400 struct btrfs_async_commit
*ac
;
1401 struct btrfs_transaction
*cur_trans
;
1403 ac
= kmalloc(sizeof(*ac
), GFP_NOFS
);
1407 INIT_WORK(&ac
->work
, do_async_commit
);
1409 ac
->newtrans
= btrfs_join_transaction(root
);
1410 if (IS_ERR(ac
->newtrans
)) {
1411 int err
= PTR_ERR(ac
->newtrans
);
1416 /* take transaction reference */
1417 cur_trans
= trans
->transaction
;
1418 atomic_inc(&cur_trans
->use_count
);
1420 btrfs_end_transaction(trans
, root
);
1423 * Tell lockdep we've released the freeze rwsem, since the
1424 * async commit thread will be the one to unlock it.
1426 if (trans
->type
< TRANS_JOIN_NOLOCK
)
1428 &root
->fs_info
->sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
1431 schedule_work(&ac
->work
);
1433 /* wait for transaction to start and unblock */
1434 if (wait_for_unblock
)
1435 wait_current_trans_commit_start_and_unblock(root
, cur_trans
);
1437 wait_current_trans_commit_start(root
, cur_trans
);
1439 if (current
->journal_info
== trans
)
1440 current
->journal_info
= NULL
;
1442 put_transaction(cur_trans
);
1447 static void cleanup_transaction(struct btrfs_trans_handle
*trans
,
1448 struct btrfs_root
*root
, int err
)
1450 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1453 WARN_ON(trans
->use_count
> 1);
1455 btrfs_abort_transaction(trans
, root
, err
);
1457 spin_lock(&root
->fs_info
->trans_lock
);
1459 if (list_empty(&cur_trans
->list
)) {
1460 spin_unlock(&root
->fs_info
->trans_lock
);
1461 btrfs_end_transaction(trans
, root
);
1465 list_del_init(&cur_trans
->list
);
1466 if (cur_trans
== root
->fs_info
->running_transaction
) {
1467 root
->fs_info
->trans_no_join
= 1;
1468 spin_unlock(&root
->fs_info
->trans_lock
);
1469 wait_event(cur_trans
->writer_wait
,
1470 atomic_read(&cur_trans
->num_writers
) == 1);
1472 spin_lock(&root
->fs_info
->trans_lock
);
1473 root
->fs_info
->running_transaction
= NULL
;
1475 spin_unlock(&root
->fs_info
->trans_lock
);
1477 btrfs_cleanup_one_transaction(trans
->transaction
, root
);
1479 put_transaction(cur_trans
);
1480 put_transaction(cur_trans
);
1482 trace_btrfs_transaction_commit(root
);
1484 btrfs_scrub_continue(root
);
1486 if (current
->journal_info
== trans
)
1487 current
->journal_info
= NULL
;
1489 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1492 static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle
*trans
,
1493 struct btrfs_root
*root
)
1495 int flush_on_commit
= btrfs_test_opt(root
, FLUSHONCOMMIT
);
1496 int snap_pending
= 0;
1499 if (!flush_on_commit
) {
1500 spin_lock(&root
->fs_info
->trans_lock
);
1501 if (!list_empty(&trans
->transaction
->pending_snapshots
))
1503 spin_unlock(&root
->fs_info
->trans_lock
);
1506 if (flush_on_commit
|| snap_pending
) {
1507 ret
= btrfs_start_delalloc_inodes(root
, 1);
1510 btrfs_wait_ordered_extents(root
, 1);
1513 ret
= btrfs_run_delayed_items(trans
, root
);
1518 * running the delayed items may have added new refs. account
1519 * them now so that they hinder processing of more delayed refs
1520 * as little as possible.
1522 btrfs_delayed_refs_qgroup_accounting(trans
, root
->fs_info
);
1525 * rename don't use btrfs_join_transaction, so, once we
1526 * set the transaction to blocked above, we aren't going
1527 * to get any new ordered operations. We can safely run
1528 * it here and no for sure that nothing new will be added
1531 ret
= btrfs_run_ordered_operations(trans
, root
, 1);
1537 * btrfs_transaction state sequence:
1538 * in_commit = 0, blocked = 0 (initial)
1539 * in_commit = 1, blocked = 1
1543 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
,
1544 struct btrfs_root
*root
)
1546 unsigned long joined
= 0;
1547 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1548 struct btrfs_transaction
*prev_trans
= NULL
;
1551 int should_grow
= 0;
1552 unsigned long now
= get_seconds();
1554 ret
= btrfs_run_ordered_operations(trans
, root
, 0);
1556 btrfs_abort_transaction(trans
, root
, ret
);
1557 btrfs_end_transaction(trans
, root
);
1561 /* Stop the commit early if ->aborted is set */
1562 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1563 ret
= cur_trans
->aborted
;
1564 btrfs_end_transaction(trans
, root
);
1568 /* make a pass through all the delayed refs we have so far
1569 * any runnings procs may add more while we are here
1571 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1573 btrfs_end_transaction(trans
, root
);
1577 btrfs_trans_release_metadata(trans
, root
);
1578 trans
->block_rsv
= NULL
;
1579 if (trans
->qgroup_reserved
) {
1580 btrfs_qgroup_free(root
, trans
->qgroup_reserved
);
1581 trans
->qgroup_reserved
= 0;
1584 cur_trans
= trans
->transaction
;
1587 * set the flushing flag so procs in this transaction have to
1588 * start sending their work down.
1590 cur_trans
->delayed_refs
.flushing
= 1;
1592 if (!list_empty(&trans
->new_bgs
))
1593 btrfs_create_pending_block_groups(trans
, root
);
1595 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1597 btrfs_end_transaction(trans
, root
);
1601 spin_lock(&cur_trans
->commit_lock
);
1602 if (cur_trans
->in_commit
) {
1603 spin_unlock(&cur_trans
->commit_lock
);
1604 atomic_inc(&cur_trans
->use_count
);
1605 ret
= btrfs_end_transaction(trans
, root
);
1607 wait_for_commit(root
, cur_trans
);
1609 put_transaction(cur_trans
);
1614 trans
->transaction
->in_commit
= 1;
1615 trans
->transaction
->blocked
= 1;
1616 spin_unlock(&cur_trans
->commit_lock
);
1617 wake_up(&root
->fs_info
->transaction_blocked_wait
);
1619 spin_lock(&root
->fs_info
->trans_lock
);
1620 if (cur_trans
->list
.prev
!= &root
->fs_info
->trans_list
) {
1621 prev_trans
= list_entry(cur_trans
->list
.prev
,
1622 struct btrfs_transaction
, list
);
1623 if (!prev_trans
->commit_done
) {
1624 atomic_inc(&prev_trans
->use_count
);
1625 spin_unlock(&root
->fs_info
->trans_lock
);
1627 wait_for_commit(root
, prev_trans
);
1629 put_transaction(prev_trans
);
1631 spin_unlock(&root
->fs_info
->trans_lock
);
1634 spin_unlock(&root
->fs_info
->trans_lock
);
1637 if (!btrfs_test_opt(root
, SSD
) &&
1638 (now
< cur_trans
->start_time
|| now
- cur_trans
->start_time
< 1))
1642 joined
= cur_trans
->num_joined
;
1644 WARN_ON(cur_trans
!= trans
->transaction
);
1646 ret
= btrfs_flush_all_pending_stuffs(trans
, root
);
1648 goto cleanup_transaction
;
1650 prepare_to_wait(&cur_trans
->writer_wait
, &wait
,
1651 TASK_UNINTERRUPTIBLE
);
1653 if (atomic_read(&cur_trans
->num_writers
) > 1)
1654 schedule_timeout(MAX_SCHEDULE_TIMEOUT
);
1655 else if (should_grow
)
1656 schedule_timeout(1);
1658 finish_wait(&cur_trans
->writer_wait
, &wait
);
1659 } while (atomic_read(&cur_trans
->num_writers
) > 1 ||
1660 (should_grow
&& cur_trans
->num_joined
!= joined
));
1662 ret
= btrfs_flush_all_pending_stuffs(trans
, root
);
1664 goto cleanup_transaction
;
1667 * Ok now we need to make sure to block out any other joins while we
1668 * commit the transaction. We could have started a join before setting
1669 * no_join so make sure to wait for num_writers to == 1 again.
1671 spin_lock(&root
->fs_info
->trans_lock
);
1672 root
->fs_info
->trans_no_join
= 1;
1673 spin_unlock(&root
->fs_info
->trans_lock
);
1674 wait_event(cur_trans
->writer_wait
,
1675 atomic_read(&cur_trans
->num_writers
) == 1);
1677 /* ->aborted might be set after the previous check, so check it */
1678 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1679 ret
= cur_trans
->aborted
;
1680 goto cleanup_transaction
;
1683 * the reloc mutex makes sure that we stop
1684 * the balancing code from coming in and moving
1685 * extents around in the middle of the commit
1687 mutex_lock(&root
->fs_info
->reloc_mutex
);
1690 * We needn't worry about the delayed items because we will
1691 * deal with them in create_pending_snapshot(), which is the
1692 * core function of the snapshot creation.
1694 ret
= create_pending_snapshots(trans
, root
->fs_info
);
1696 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1697 goto cleanup_transaction
;
1701 * We insert the dir indexes of the snapshots and update the inode
1702 * of the snapshots' parents after the snapshot creation, so there
1703 * are some delayed items which are not dealt with. Now deal with
1706 * We needn't worry that this operation will corrupt the snapshots,
1707 * because all the tree which are snapshoted will be forced to COW
1708 * the nodes and leaves.
1710 ret
= btrfs_run_delayed_items(trans
, root
);
1712 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1713 goto cleanup_transaction
;
1716 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1718 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1719 goto cleanup_transaction
;
1723 * make sure none of the code above managed to slip in a
1726 btrfs_assert_delayed_root_empty(root
);
1728 WARN_ON(cur_trans
!= trans
->transaction
);
1730 btrfs_scrub_pause(root
);
1731 /* btrfs_commit_tree_roots is responsible for getting the
1732 * various roots consistent with each other. Every pointer
1733 * in the tree of tree roots has to point to the most up to date
1734 * root for every subvolume and other tree. So, we have to keep
1735 * the tree logging code from jumping in and changing any
1738 * At this point in the commit, there can't be any tree-log
1739 * writers, but a little lower down we drop the trans mutex
1740 * and let new people in. By holding the tree_log_mutex
1741 * from now until after the super is written, we avoid races
1742 * with the tree-log code.
1744 mutex_lock(&root
->fs_info
->tree_log_mutex
);
1746 ret
= commit_fs_roots(trans
, root
);
1748 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1749 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1750 goto cleanup_transaction
;
1753 /* commit_fs_roots gets rid of all the tree log roots, it is now
1754 * safe to free the root of tree log roots
1756 btrfs_free_log_root_tree(trans
, root
->fs_info
);
1758 ret
= commit_cowonly_roots(trans
, root
);
1760 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1761 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1762 goto cleanup_transaction
;
1766 * The tasks which save the space cache and inode cache may also
1767 * update ->aborted, check it.
1769 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1770 ret
= cur_trans
->aborted
;
1771 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1772 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1773 goto cleanup_transaction
;
1776 btrfs_prepare_extent_commit(trans
, root
);
1778 cur_trans
= root
->fs_info
->running_transaction
;
1780 btrfs_set_root_node(&root
->fs_info
->tree_root
->root_item
,
1781 root
->fs_info
->tree_root
->node
);
1782 switch_commit_root(root
->fs_info
->tree_root
);
1784 btrfs_set_root_node(&root
->fs_info
->chunk_root
->root_item
,
1785 root
->fs_info
->chunk_root
->node
);
1786 switch_commit_root(root
->fs_info
->chunk_root
);
1788 assert_qgroups_uptodate(trans
);
1789 update_super_roots(root
);
1791 if (!root
->fs_info
->log_root_recovering
) {
1792 btrfs_set_super_log_root(root
->fs_info
->super_copy
, 0);
1793 btrfs_set_super_log_root_level(root
->fs_info
->super_copy
, 0);
1796 memcpy(root
->fs_info
->super_for_commit
, root
->fs_info
->super_copy
,
1797 sizeof(*root
->fs_info
->super_copy
));
1799 trans
->transaction
->blocked
= 0;
1800 spin_lock(&root
->fs_info
->trans_lock
);
1801 root
->fs_info
->running_transaction
= NULL
;
1802 root
->fs_info
->trans_no_join
= 0;
1803 spin_unlock(&root
->fs_info
->trans_lock
);
1804 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1806 wake_up(&root
->fs_info
->transaction_wait
);
1808 ret
= btrfs_write_and_wait_transaction(trans
, root
);
1810 btrfs_error(root
->fs_info
, ret
,
1811 "Error while writing out transaction.");
1812 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1813 goto cleanup_transaction
;
1816 ret
= write_ctree_super(trans
, root
, 0);
1818 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1819 goto cleanup_transaction
;
1823 * the super is written, we can safely allow the tree-loggers
1824 * to go about their business
1826 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1828 btrfs_finish_extent_commit(trans
, root
);
1830 cur_trans
->commit_done
= 1;
1832 root
->fs_info
->last_trans_committed
= cur_trans
->transid
;
1834 wake_up(&cur_trans
->commit_wait
);
1836 spin_lock(&root
->fs_info
->trans_lock
);
1837 list_del_init(&cur_trans
->list
);
1838 spin_unlock(&root
->fs_info
->trans_lock
);
1840 put_transaction(cur_trans
);
1841 put_transaction(cur_trans
);
1843 if (trans
->type
< TRANS_JOIN_NOLOCK
)
1844 sb_end_intwrite(root
->fs_info
->sb
);
1846 trace_btrfs_transaction_commit(root
);
1848 btrfs_scrub_continue(root
);
1850 if (current
->journal_info
== trans
)
1851 current
->journal_info
= NULL
;
1853 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1855 if (current
!= root
->fs_info
->transaction_kthread
)
1856 btrfs_run_delayed_iputs(root
);
1860 cleanup_transaction
:
1861 btrfs_trans_release_metadata(trans
, root
);
1862 trans
->block_rsv
= NULL
;
1863 if (trans
->qgroup_reserved
) {
1864 btrfs_qgroup_free(root
, trans
->qgroup_reserved
);
1865 trans
->qgroup_reserved
= 0;
1867 btrfs_printk(root
->fs_info
, "Skipping commit of aborted transaction.\n");
1869 if (current
->journal_info
== trans
)
1870 current
->journal_info
= NULL
;
1871 cleanup_transaction(trans
, root
, ret
);
1877 * interface function to delete all the snapshots we have scheduled for deletion
1879 int btrfs_clean_old_snapshots(struct btrfs_root
*root
)
1882 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1884 spin_lock(&fs_info
->trans_lock
);
1885 list_splice_init(&fs_info
->dead_roots
, &list
);
1886 spin_unlock(&fs_info
->trans_lock
);
1888 while (!list_empty(&list
)) {
1891 root
= list_entry(list
.next
, struct btrfs_root
, root_list
);
1892 list_del(&root
->root_list
);
1894 btrfs_kill_all_delayed_nodes(root
);
1896 if (btrfs_header_backref_rev(root
->node
) <
1897 BTRFS_MIXED_BACKREF_REV
)
1898 ret
= btrfs_drop_snapshot(root
, NULL
, 0, 0);
1900 ret
=btrfs_drop_snapshot(root
, NULL
, 1, 0);