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 static 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
&&
57 (type
& TRANS_EXTWRITERS
));
60 static inline void extwriter_counter_inc(struct btrfs_transaction
*trans
,
63 if (type
& TRANS_EXTWRITERS
)
64 atomic_inc(&trans
->num_extwriters
);
67 static inline void extwriter_counter_dec(struct btrfs_transaction
*trans
,
70 if (type
& TRANS_EXTWRITERS
)
71 atomic_dec(&trans
->num_extwriters
);
74 static inline void extwriter_counter_init(struct btrfs_transaction
*trans
,
77 atomic_set(&trans
->num_extwriters
, ((type
& TRANS_EXTWRITERS
) ? 1 : 0));
80 static inline int extwriter_counter_read(struct btrfs_transaction
*trans
)
82 return atomic_read(&trans
->num_extwriters
);
86 * either allocate a new transaction or hop into the existing one
88 static noinline
int join_transaction(struct btrfs_root
*root
, unsigned int type
)
90 struct btrfs_transaction
*cur_trans
;
91 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
93 spin_lock(&fs_info
->trans_lock
);
95 /* The file system has been taken offline. No new transactions. */
96 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
97 spin_unlock(&fs_info
->trans_lock
);
101 if (fs_info
->trans_no_join
) {
103 * If we are JOIN_NOLOCK we're already committing a current
104 * transaction, we just need a handle to deal with something
105 * when committing the transaction, such as inode cache and
106 * space cache. It is a special case.
108 if (type
!= TRANS_JOIN_NOLOCK
) {
109 spin_unlock(&fs_info
->trans_lock
);
114 cur_trans
= fs_info
->running_transaction
;
116 if (cur_trans
->aborted
) {
117 spin_unlock(&fs_info
->trans_lock
);
118 return cur_trans
->aborted
;
120 if (!can_join_transaction(cur_trans
, type
)) {
121 spin_unlock(&fs_info
->trans_lock
);
124 atomic_inc(&cur_trans
->use_count
);
125 atomic_inc(&cur_trans
->num_writers
);
126 extwriter_counter_inc(cur_trans
, type
);
127 cur_trans
->num_joined
++;
128 spin_unlock(&fs_info
->trans_lock
);
131 spin_unlock(&fs_info
->trans_lock
);
134 * If we are ATTACH, we just want to catch the current transaction,
135 * and commit it. If there is no transaction, just return ENOENT.
137 if (type
== TRANS_ATTACH
)
140 cur_trans
= kmem_cache_alloc(btrfs_transaction_cachep
, GFP_NOFS
);
144 spin_lock(&fs_info
->trans_lock
);
145 if (fs_info
->running_transaction
) {
147 * someone started a transaction after we unlocked. Make sure
148 * to redo the trans_no_join checks above
150 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
152 } else if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
153 spin_unlock(&fs_info
->trans_lock
);
154 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
158 atomic_set(&cur_trans
->num_writers
, 1);
159 extwriter_counter_init(cur_trans
, type
);
160 cur_trans
->num_joined
= 0;
161 init_waitqueue_head(&cur_trans
->writer_wait
);
162 init_waitqueue_head(&cur_trans
->commit_wait
);
163 cur_trans
->in_commit
= 0;
164 cur_trans
->blocked
= 0;
166 * One for this trans handle, one so it will live on until we
167 * commit the transaction.
169 atomic_set(&cur_trans
->use_count
, 2);
170 cur_trans
->commit_done
= 0;
171 cur_trans
->start_time
= get_seconds();
173 cur_trans
->delayed_refs
.root
= RB_ROOT
;
174 cur_trans
->delayed_refs
.num_entries
= 0;
175 cur_trans
->delayed_refs
.num_heads_ready
= 0;
176 cur_trans
->delayed_refs
.num_heads
= 0;
177 cur_trans
->delayed_refs
.flushing
= 0;
178 cur_trans
->delayed_refs
.run_delayed_start
= 0;
181 * although the tree mod log is per file system and not per transaction,
182 * the log must never go across transaction boundaries.
185 if (!list_empty(&fs_info
->tree_mod_seq_list
))
186 WARN(1, KERN_ERR
"btrfs: tree_mod_seq_list not empty when "
187 "creating a fresh transaction\n");
188 if (!RB_EMPTY_ROOT(&fs_info
->tree_mod_log
))
189 WARN(1, KERN_ERR
"btrfs: tree_mod_log rb tree not empty when "
190 "creating a fresh transaction\n");
191 atomic64_set(&fs_info
->tree_mod_seq
, 0);
193 spin_lock_init(&cur_trans
->commit_lock
);
194 spin_lock_init(&cur_trans
->delayed_refs
.lock
);
195 atomic_set(&cur_trans
->delayed_refs
.procs_running_refs
, 0);
196 atomic_set(&cur_trans
->delayed_refs
.ref_seq
, 0);
197 init_waitqueue_head(&cur_trans
->delayed_refs
.wait
);
199 INIT_LIST_HEAD(&cur_trans
->pending_snapshots
);
200 INIT_LIST_HEAD(&cur_trans
->ordered_operations
);
201 list_add_tail(&cur_trans
->list
, &fs_info
->trans_list
);
202 extent_io_tree_init(&cur_trans
->dirty_pages
,
203 fs_info
->btree_inode
->i_mapping
);
204 fs_info
->generation
++;
205 cur_trans
->transid
= fs_info
->generation
;
206 fs_info
->running_transaction
= cur_trans
;
207 cur_trans
->aborted
= 0;
208 spin_unlock(&fs_info
->trans_lock
);
214 * this does all the record keeping required to make sure that a reference
215 * counted root is properly recorded in a given transaction. This is required
216 * to make sure the old root from before we joined the transaction is deleted
217 * when the transaction commits
219 static int record_root_in_trans(struct btrfs_trans_handle
*trans
,
220 struct btrfs_root
*root
)
222 if (root
->ref_cows
&& root
->last_trans
< trans
->transid
) {
223 WARN_ON(root
== root
->fs_info
->extent_root
);
224 WARN_ON(root
->commit_root
!= root
->node
);
227 * see below for in_trans_setup usage rules
228 * we have the reloc mutex held now, so there
229 * is only one writer in this function
231 root
->in_trans_setup
= 1;
233 /* make sure readers find in_trans_setup before
234 * they find our root->last_trans update
238 spin_lock(&root
->fs_info
->fs_roots_radix_lock
);
239 if (root
->last_trans
== trans
->transid
) {
240 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
243 radix_tree_tag_set(&root
->fs_info
->fs_roots_radix
,
244 (unsigned long)root
->root_key
.objectid
,
245 BTRFS_ROOT_TRANS_TAG
);
246 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
247 root
->last_trans
= trans
->transid
;
249 /* this is pretty tricky. We don't want to
250 * take the relocation lock in btrfs_record_root_in_trans
251 * unless we're really doing the first setup for this root in
254 * Normally we'd use root->last_trans as a flag to decide
255 * if we want to take the expensive mutex.
257 * But, we have to set root->last_trans before we
258 * init the relocation root, otherwise, we trip over warnings
259 * in ctree.c. The solution used here is to flag ourselves
260 * with root->in_trans_setup. When this is 1, we're still
261 * fixing up the reloc trees and everyone must wait.
263 * When this is zero, they can trust root->last_trans and fly
264 * through btrfs_record_root_in_trans without having to take the
265 * lock. smp_wmb() makes sure that all the writes above are
266 * done before we pop in the zero below
268 btrfs_init_reloc_root(trans
, root
);
270 root
->in_trans_setup
= 0;
276 int btrfs_record_root_in_trans(struct btrfs_trans_handle
*trans
,
277 struct btrfs_root
*root
)
283 * see record_root_in_trans for comments about in_trans_setup usage
287 if (root
->last_trans
== trans
->transid
&&
288 !root
->in_trans_setup
)
291 mutex_lock(&root
->fs_info
->reloc_mutex
);
292 record_root_in_trans(trans
, root
);
293 mutex_unlock(&root
->fs_info
->reloc_mutex
);
298 /* wait for commit against the current transaction to become unblocked
299 * when this is done, it is safe to start a new transaction, but the current
300 * transaction might not be fully on disk.
302 static void wait_current_trans(struct btrfs_root
*root
)
304 struct btrfs_transaction
*cur_trans
;
306 spin_lock(&root
->fs_info
->trans_lock
);
307 cur_trans
= root
->fs_info
->running_transaction
;
308 if (cur_trans
&& cur_trans
->blocked
) {
309 atomic_inc(&cur_trans
->use_count
);
310 spin_unlock(&root
->fs_info
->trans_lock
);
312 wait_event(root
->fs_info
->transaction_wait
,
313 !cur_trans
->blocked
);
314 put_transaction(cur_trans
);
316 spin_unlock(&root
->fs_info
->trans_lock
);
320 static int may_wait_transaction(struct btrfs_root
*root
, int type
)
322 if (root
->fs_info
->log_root_recovering
)
325 if (type
== TRANS_USERSPACE
)
328 if (type
== TRANS_START
&&
329 !atomic_read(&root
->fs_info
->open_ioctl_trans
))
335 static struct btrfs_trans_handle
*
336 start_transaction(struct btrfs_root
*root
, u64 num_items
, unsigned int type
,
337 enum btrfs_reserve_flush_enum flush
)
339 struct btrfs_trans_handle
*h
;
340 struct btrfs_transaction
*cur_trans
;
343 u64 qgroup_reserved
= 0;
345 if (test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
346 return ERR_PTR(-EROFS
);
348 if (current
->journal_info
) {
349 WARN_ON(type
& TRANS_EXTWRITERS
);
350 h
= current
->journal_info
;
352 WARN_ON(h
->use_count
> 2);
353 h
->orig_rsv
= h
->block_rsv
;
359 * Do the reservation before we join the transaction so we can do all
360 * the appropriate flushing if need be.
362 if (num_items
> 0 && root
!= root
->fs_info
->chunk_root
) {
363 if (root
->fs_info
->quota_enabled
&&
364 is_fstree(root
->root_key
.objectid
)) {
365 qgroup_reserved
= num_items
* root
->leafsize
;
366 ret
= btrfs_qgroup_reserve(root
, qgroup_reserved
);
371 num_bytes
= btrfs_calc_trans_metadata_size(root
, num_items
);
372 ret
= btrfs_block_rsv_add(root
,
373 &root
->fs_info
->trans_block_rsv
,
379 h
= kmem_cache_alloc(btrfs_trans_handle_cachep
, GFP_NOFS
);
386 * If we are JOIN_NOLOCK we're already committing a transaction and
387 * waiting on this guy, so we don't need to do the sb_start_intwrite
388 * because we're already holding a ref. We need this because we could
389 * have raced in and did an fsync() on a file which can kick a commit
390 * and then we deadlock with somebody doing a freeze.
392 * If we are ATTACH, it means we just want to catch the current
393 * transaction and commit it, so we needn't do sb_start_intwrite().
395 if (type
& __TRANS_FREEZABLE
)
396 sb_start_intwrite(root
->fs_info
->sb
);
398 if (may_wait_transaction(root
, type
))
399 wait_current_trans(root
);
402 ret
= join_transaction(root
, type
);
404 wait_current_trans(root
);
405 if (unlikely(type
== TRANS_ATTACH
))
408 } while (ret
== -EBUSY
);
411 /* We must get the transaction if we are JOIN_NOLOCK. */
412 BUG_ON(type
== TRANS_JOIN_NOLOCK
);
416 cur_trans
= root
->fs_info
->running_transaction
;
418 h
->transid
= cur_trans
->transid
;
419 h
->transaction
= cur_trans
;
421 h
->bytes_reserved
= 0;
423 h
->delayed_ref_updates
= 0;
429 h
->qgroup_reserved
= 0;
430 h
->delayed_ref_elem
.seq
= 0;
432 h
->allocating_chunk
= false;
433 INIT_LIST_HEAD(&h
->qgroup_ref_list
);
434 INIT_LIST_HEAD(&h
->new_bgs
);
437 if (cur_trans
->blocked
&& may_wait_transaction(root
, type
)) {
438 btrfs_commit_transaction(h
, root
);
443 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
444 h
->transid
, num_bytes
, 1);
445 h
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
446 h
->bytes_reserved
= num_bytes
;
448 h
->qgroup_reserved
= qgroup_reserved
;
451 btrfs_record_root_in_trans(h
, root
);
453 if (!current
->journal_info
&& type
!= TRANS_USERSPACE
)
454 current
->journal_info
= h
;
458 if (type
& __TRANS_FREEZABLE
)
459 sb_end_intwrite(root
->fs_info
->sb
);
460 kmem_cache_free(btrfs_trans_handle_cachep
, h
);
463 btrfs_block_rsv_release(root
, &root
->fs_info
->trans_block_rsv
,
467 btrfs_qgroup_free(root
, qgroup_reserved
);
471 struct btrfs_trans_handle
*btrfs_start_transaction(struct btrfs_root
*root
,
474 return start_transaction(root
, num_items
, TRANS_START
,
475 BTRFS_RESERVE_FLUSH_ALL
);
478 struct btrfs_trans_handle
*btrfs_start_transaction_lflush(
479 struct btrfs_root
*root
, int num_items
)
481 return start_transaction(root
, num_items
, TRANS_START
,
482 BTRFS_RESERVE_FLUSH_LIMIT
);
485 struct btrfs_trans_handle
*btrfs_join_transaction(struct btrfs_root
*root
)
487 return start_transaction(root
, 0, TRANS_JOIN
, 0);
490 struct btrfs_trans_handle
*btrfs_join_transaction_nolock(struct btrfs_root
*root
)
492 return start_transaction(root
, 0, TRANS_JOIN_NOLOCK
, 0);
495 struct btrfs_trans_handle
*btrfs_start_ioctl_transaction(struct btrfs_root
*root
)
497 return start_transaction(root
, 0, TRANS_USERSPACE
, 0);
501 * btrfs_attach_transaction() - catch the running transaction
503 * It is used when we want to commit the current the transaction, but
504 * don't want to start a new one.
506 * Note: If this function return -ENOENT, it just means there is no
507 * running transaction. But it is possible that the inactive transaction
508 * is still in the memory, not fully on disk. If you hope there is no
509 * inactive transaction in the fs when -ENOENT is returned, you should
511 * btrfs_attach_transaction_barrier()
513 struct btrfs_trans_handle
*btrfs_attach_transaction(struct btrfs_root
*root
)
515 return start_transaction(root
, 0, TRANS_ATTACH
, 0);
519 * btrfs_attach_transaction() - catch the running transaction
521 * It is similar to the above function, the differentia is this one
522 * will wait for all the inactive transactions until they fully
525 struct btrfs_trans_handle
*
526 btrfs_attach_transaction_barrier(struct btrfs_root
*root
)
528 struct btrfs_trans_handle
*trans
;
530 trans
= start_transaction(root
, 0, TRANS_ATTACH
, 0);
531 if (IS_ERR(trans
) && PTR_ERR(trans
) == -ENOENT
)
532 btrfs_wait_for_commit(root
, 0);
537 /* wait for a transaction commit to be fully complete */
538 static noinline
void wait_for_commit(struct btrfs_root
*root
,
539 struct btrfs_transaction
*commit
)
541 wait_event(commit
->commit_wait
, commit
->commit_done
);
544 int btrfs_wait_for_commit(struct btrfs_root
*root
, u64 transid
)
546 struct btrfs_transaction
*cur_trans
= NULL
, *t
;
550 if (transid
<= root
->fs_info
->last_trans_committed
)
554 /* find specified transaction */
555 spin_lock(&root
->fs_info
->trans_lock
);
556 list_for_each_entry(t
, &root
->fs_info
->trans_list
, list
) {
557 if (t
->transid
== transid
) {
559 atomic_inc(&cur_trans
->use_count
);
563 if (t
->transid
> transid
) {
568 spin_unlock(&root
->fs_info
->trans_lock
);
569 /* The specified transaction doesn't exist */
573 /* find newest transaction that is committing | committed */
574 spin_lock(&root
->fs_info
->trans_lock
);
575 list_for_each_entry_reverse(t
, &root
->fs_info
->trans_list
,
581 atomic_inc(&cur_trans
->use_count
);
585 spin_unlock(&root
->fs_info
->trans_lock
);
587 goto out
; /* nothing committing|committed */
590 wait_for_commit(root
, cur_trans
);
591 put_transaction(cur_trans
);
596 void btrfs_throttle(struct btrfs_root
*root
)
598 if (!atomic_read(&root
->fs_info
->open_ioctl_trans
))
599 wait_current_trans(root
);
602 static int should_end_transaction(struct btrfs_trans_handle
*trans
,
603 struct btrfs_root
*root
)
607 ret
= btrfs_block_rsv_check(root
, &root
->fs_info
->global_block_rsv
, 5);
611 int btrfs_should_end_transaction(struct btrfs_trans_handle
*trans
,
612 struct btrfs_root
*root
)
614 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
619 if (cur_trans
->blocked
|| cur_trans
->delayed_refs
.flushing
)
622 updates
= trans
->delayed_ref_updates
;
623 trans
->delayed_ref_updates
= 0;
625 err
= btrfs_run_delayed_refs(trans
, root
, updates
);
626 if (err
) /* Error code will also eval true */
630 return should_end_transaction(trans
, root
);
633 static int __btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
634 struct btrfs_root
*root
, int throttle
)
636 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
637 struct btrfs_fs_info
*info
= root
->fs_info
;
639 int lock
= (trans
->type
!= TRANS_JOIN_NOLOCK
);
642 if (--trans
->use_count
) {
643 trans
->block_rsv
= trans
->orig_rsv
;
648 * do the qgroup accounting as early as possible
650 err
= btrfs_delayed_refs_qgroup_accounting(trans
, info
);
652 btrfs_trans_release_metadata(trans
, root
);
653 trans
->block_rsv
= NULL
;
655 if (trans
->qgroup_reserved
) {
657 * the same root has to be passed here between start_transaction
658 * and end_transaction. Subvolume quota depends on this.
660 btrfs_qgroup_free(trans
->root
, trans
->qgroup_reserved
);
661 trans
->qgroup_reserved
= 0;
664 if (!list_empty(&trans
->new_bgs
))
665 btrfs_create_pending_block_groups(trans
, root
);
668 unsigned long cur
= trans
->delayed_ref_updates
;
669 trans
->delayed_ref_updates
= 0;
671 trans
->transaction
->delayed_refs
.num_heads_ready
> 64) {
672 trans
->delayed_ref_updates
= 0;
673 btrfs_run_delayed_refs(trans
, root
, cur
);
680 btrfs_trans_release_metadata(trans
, root
);
681 trans
->block_rsv
= NULL
;
683 if (!list_empty(&trans
->new_bgs
))
684 btrfs_create_pending_block_groups(trans
, root
);
686 if (lock
&& !atomic_read(&root
->fs_info
->open_ioctl_trans
) &&
687 should_end_transaction(trans
, root
)) {
688 trans
->transaction
->blocked
= 1;
692 if (lock
&& cur_trans
->blocked
&& !cur_trans
->in_commit
) {
695 * We may race with somebody else here so end up having
696 * to call end_transaction on ourselves again, so inc
700 return btrfs_commit_transaction(trans
, root
);
702 wake_up_process(info
->transaction_kthread
);
706 if (trans
->type
& __TRANS_FREEZABLE
)
707 sb_end_intwrite(root
->fs_info
->sb
);
709 WARN_ON(cur_trans
!= info
->running_transaction
);
710 WARN_ON(atomic_read(&cur_trans
->num_writers
) < 1);
711 atomic_dec(&cur_trans
->num_writers
);
712 extwriter_counter_dec(cur_trans
, trans
->type
);
715 if (waitqueue_active(&cur_trans
->writer_wait
))
716 wake_up(&cur_trans
->writer_wait
);
717 put_transaction(cur_trans
);
719 if (current
->journal_info
== trans
)
720 current
->journal_info
= NULL
;
723 btrfs_run_delayed_iputs(root
);
725 if (trans
->aborted
||
726 test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
728 assert_qgroups_uptodate(trans
);
730 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
734 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
735 struct btrfs_root
*root
)
737 return __btrfs_end_transaction(trans
, root
, 0);
740 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
,
741 struct btrfs_root
*root
)
743 return __btrfs_end_transaction(trans
, root
, 1);
746 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle
*trans
,
747 struct btrfs_root
*root
)
749 return __btrfs_end_transaction(trans
, root
, 1);
753 * when btree blocks are allocated, they have some corresponding bits set for
754 * them in one of two extent_io trees. This is used to make sure all of
755 * those extents are sent to disk but does not wait on them
757 int btrfs_write_marked_extents(struct btrfs_root
*root
,
758 struct extent_io_tree
*dirty_pages
, int mark
)
762 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
763 struct extent_state
*cached_state
= NULL
;
766 struct blk_plug plug
;
768 blk_start_plug(&plug
);
769 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
770 mark
, &cached_state
)) {
771 convert_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
,
772 mark
, &cached_state
, GFP_NOFS
);
774 err
= filemap_fdatawrite_range(mapping
, start
, end
);
782 blk_finish_plug(&plug
);
787 * when btree blocks are allocated, they have some corresponding bits set for
788 * them in one of two extent_io trees. This is used to make sure all of
789 * those extents are on disk for transaction or log commit. We wait
790 * on all the pages and clear them from the dirty pages state tree
792 int btrfs_wait_marked_extents(struct btrfs_root
*root
,
793 struct extent_io_tree
*dirty_pages
, int mark
)
797 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
798 struct extent_state
*cached_state
= NULL
;
802 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
803 EXTENT_NEED_WAIT
, &cached_state
)) {
804 clear_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
,
805 0, 0, &cached_state
, GFP_NOFS
);
806 err
= filemap_fdatawait_range(mapping
, start
, end
);
818 * when btree blocks are allocated, they have some corresponding bits set for
819 * them in one of two extent_io trees. This is used to make sure all of
820 * those extents are on disk for transaction or log commit
822 int btrfs_write_and_wait_marked_extents(struct btrfs_root
*root
,
823 struct extent_io_tree
*dirty_pages
, int mark
)
828 ret
= btrfs_write_marked_extents(root
, dirty_pages
, mark
);
829 ret2
= btrfs_wait_marked_extents(root
, dirty_pages
, mark
);
838 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
,
839 struct btrfs_root
*root
)
841 if (!trans
|| !trans
->transaction
) {
842 struct inode
*btree_inode
;
843 btree_inode
= root
->fs_info
->btree_inode
;
844 return filemap_write_and_wait(btree_inode
->i_mapping
);
846 return btrfs_write_and_wait_marked_extents(root
,
847 &trans
->transaction
->dirty_pages
,
852 * this is used to update the root pointer in the tree of tree roots.
854 * But, in the case of the extent allocation tree, updating the root
855 * pointer may allocate blocks which may change the root of the extent
858 * So, this loops and repeats and makes sure the cowonly root didn't
859 * change while the root pointer was being updated in the metadata.
861 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
862 struct btrfs_root
*root
)
867 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
869 old_root_used
= btrfs_root_used(&root
->root_item
);
870 btrfs_write_dirty_block_groups(trans
, root
);
873 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
874 if (old_root_bytenr
== root
->node
->start
&&
875 old_root_used
== btrfs_root_used(&root
->root_item
))
878 btrfs_set_root_node(&root
->root_item
, root
->node
);
879 ret
= btrfs_update_root(trans
, tree_root
,
885 old_root_used
= btrfs_root_used(&root
->root_item
);
886 ret
= btrfs_write_dirty_block_groups(trans
, root
);
891 if (root
!= root
->fs_info
->extent_root
)
892 switch_commit_root(root
);
898 * update all the cowonly tree roots on disk
900 * The error handling in this function may not be obvious. Any of the
901 * failures will cause the file system to go offline. We still need
902 * to clean up the delayed refs.
904 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
,
905 struct btrfs_root
*root
)
907 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
908 struct list_head
*next
;
909 struct extent_buffer
*eb
;
912 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
916 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
917 ret
= btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
,
919 btrfs_tree_unlock(eb
);
920 free_extent_buffer(eb
);
925 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
929 ret
= btrfs_run_dev_stats(trans
, root
->fs_info
);
931 ret
= btrfs_run_dev_replace(trans
, root
->fs_info
);
934 ret
= btrfs_run_qgroups(trans
, root
->fs_info
);
937 /* run_qgroups might have added some more refs */
938 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
941 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
942 next
= fs_info
->dirty_cowonly_roots
.next
;
944 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
946 ret
= update_cowonly_root(trans
, root
);
951 down_write(&fs_info
->extent_commit_sem
);
952 switch_commit_root(fs_info
->extent_root
);
953 up_write(&fs_info
->extent_commit_sem
);
955 btrfs_after_dev_replace_commit(fs_info
);
961 * dead roots are old snapshots that need to be deleted. This allocates
962 * a dirty root struct and adds it into the list of dead roots that need to
965 int btrfs_add_dead_root(struct btrfs_root
*root
)
967 spin_lock(&root
->fs_info
->trans_lock
);
968 list_add_tail(&root
->root_list
, &root
->fs_info
->dead_roots
);
969 spin_unlock(&root
->fs_info
->trans_lock
);
974 * update all the cowonly tree roots on disk
976 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
,
977 struct btrfs_root
*root
)
979 struct btrfs_root
*gang
[8];
980 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
985 spin_lock(&fs_info
->fs_roots_radix_lock
);
987 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
990 BTRFS_ROOT_TRANS_TAG
);
993 for (i
= 0; i
< ret
; i
++) {
995 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
996 (unsigned long)root
->root_key
.objectid
,
997 BTRFS_ROOT_TRANS_TAG
);
998 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1000 btrfs_free_log(trans
, root
);
1001 btrfs_update_reloc_root(trans
, root
);
1002 btrfs_orphan_commit_root(trans
, root
);
1004 btrfs_save_ino_cache(root
, trans
);
1006 /* see comments in should_cow_block() */
1007 root
->force_cow
= 0;
1010 if (root
->commit_root
!= root
->node
) {
1011 mutex_lock(&root
->fs_commit_mutex
);
1012 switch_commit_root(root
);
1013 btrfs_unpin_free_ino(root
);
1014 mutex_unlock(&root
->fs_commit_mutex
);
1016 btrfs_set_root_node(&root
->root_item
,
1020 err
= btrfs_update_root(trans
, fs_info
->tree_root
,
1023 spin_lock(&fs_info
->fs_roots_radix_lock
);
1028 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1033 * defrag a given btree.
1034 * Every leaf in the btree is read and defragged.
1036 int btrfs_defrag_root(struct btrfs_root
*root
)
1038 struct btrfs_fs_info
*info
= root
->fs_info
;
1039 struct btrfs_trans_handle
*trans
;
1042 if (xchg(&root
->defrag_running
, 1))
1046 trans
= btrfs_start_transaction(root
, 0);
1048 return PTR_ERR(trans
);
1050 ret
= btrfs_defrag_leaves(trans
, root
);
1052 btrfs_end_transaction(trans
, root
);
1053 btrfs_btree_balance_dirty(info
->tree_root
);
1056 if (btrfs_fs_closing(root
->fs_info
) || ret
!= -EAGAIN
)
1059 if (btrfs_defrag_cancelled(root
->fs_info
)) {
1060 printk(KERN_DEBUG
"btrfs: defrag_root cancelled\n");
1065 root
->defrag_running
= 0;
1070 * new snapshots need to be created at a very specific time in the
1071 * transaction commit. This does the actual creation.
1074 * If the error which may affect the commitment of the current transaction
1075 * happens, we should return the error number. If the error which just affect
1076 * the creation of the pending snapshots, just return 0.
1078 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
1079 struct btrfs_fs_info
*fs_info
,
1080 struct btrfs_pending_snapshot
*pending
)
1082 struct btrfs_key key
;
1083 struct btrfs_root_item
*new_root_item
;
1084 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1085 struct btrfs_root
*root
= pending
->root
;
1086 struct btrfs_root
*parent_root
;
1087 struct btrfs_block_rsv
*rsv
;
1088 struct inode
*parent_inode
;
1089 struct btrfs_path
*path
;
1090 struct btrfs_dir_item
*dir_item
;
1091 struct dentry
*dentry
;
1092 struct extent_buffer
*tmp
;
1093 struct extent_buffer
*old
;
1094 struct timespec cur_time
= CURRENT_TIME
;
1102 path
= btrfs_alloc_path();
1104 pending
->error
= -ENOMEM
;
1108 new_root_item
= kmalloc(sizeof(*new_root_item
), GFP_NOFS
);
1109 if (!new_root_item
) {
1110 pending
->error
= -ENOMEM
;
1111 goto root_item_alloc_fail
;
1114 pending
->error
= btrfs_find_free_objectid(tree_root
, &objectid
);
1116 goto no_free_objectid
;
1118 btrfs_reloc_pre_snapshot(trans
, pending
, &to_reserve
);
1120 if (to_reserve
> 0) {
1121 pending
->error
= btrfs_block_rsv_add(root
,
1122 &pending
->block_rsv
,
1124 BTRFS_RESERVE_NO_FLUSH
);
1126 goto no_free_objectid
;
1129 pending
->error
= btrfs_qgroup_inherit(trans
, fs_info
,
1130 root
->root_key
.objectid
,
1131 objectid
, pending
->inherit
);
1133 goto no_free_objectid
;
1135 key
.objectid
= objectid
;
1136 key
.offset
= (u64
)-1;
1137 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1139 rsv
= trans
->block_rsv
;
1140 trans
->block_rsv
= &pending
->block_rsv
;
1141 trans
->bytes_reserved
= trans
->block_rsv
->reserved
;
1143 dentry
= pending
->dentry
;
1144 parent_inode
= pending
->dir
;
1145 parent_root
= BTRFS_I(parent_inode
)->root
;
1146 record_root_in_trans(trans
, parent_root
);
1149 * insert the directory item
1151 ret
= btrfs_set_inode_index(parent_inode
, &index
);
1152 BUG_ON(ret
); /* -ENOMEM */
1154 /* check if there is a file/dir which has the same name. */
1155 dir_item
= btrfs_lookup_dir_item(NULL
, parent_root
, path
,
1156 btrfs_ino(parent_inode
),
1157 dentry
->d_name
.name
,
1158 dentry
->d_name
.len
, 0);
1159 if (dir_item
!= NULL
&& !IS_ERR(dir_item
)) {
1160 pending
->error
= -EEXIST
;
1161 goto dir_item_existed
;
1162 } else if (IS_ERR(dir_item
)) {
1163 ret
= PTR_ERR(dir_item
);
1164 btrfs_abort_transaction(trans
, root
, ret
);
1167 btrfs_release_path(path
);
1170 * pull in the delayed directory update
1171 * and the delayed inode item
1172 * otherwise we corrupt the FS during
1175 ret
= btrfs_run_delayed_items(trans
, root
);
1176 if (ret
) { /* Transaction aborted */
1177 btrfs_abort_transaction(trans
, root
, ret
);
1181 record_root_in_trans(trans
, root
);
1182 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
1183 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
1184 btrfs_check_and_init_root_item(new_root_item
);
1186 root_flags
= btrfs_root_flags(new_root_item
);
1187 if (pending
->readonly
)
1188 root_flags
|= BTRFS_ROOT_SUBVOL_RDONLY
;
1190 root_flags
&= ~BTRFS_ROOT_SUBVOL_RDONLY
;
1191 btrfs_set_root_flags(new_root_item
, root_flags
);
1193 btrfs_set_root_generation_v2(new_root_item
,
1195 uuid_le_gen(&new_uuid
);
1196 memcpy(new_root_item
->uuid
, new_uuid
.b
, BTRFS_UUID_SIZE
);
1197 memcpy(new_root_item
->parent_uuid
, root
->root_item
.uuid
,
1199 if (!(root_flags
& BTRFS_ROOT_SUBVOL_RDONLY
)) {
1200 memset(new_root_item
->received_uuid
, 0,
1201 sizeof(new_root_item
->received_uuid
));
1202 memset(&new_root_item
->stime
, 0, sizeof(new_root_item
->stime
));
1203 memset(&new_root_item
->rtime
, 0, sizeof(new_root_item
->rtime
));
1204 btrfs_set_root_stransid(new_root_item
, 0);
1205 btrfs_set_root_rtransid(new_root_item
, 0);
1207 new_root_item
->otime
.sec
= cpu_to_le64(cur_time
.tv_sec
);
1208 new_root_item
->otime
.nsec
= cpu_to_le32(cur_time
.tv_nsec
);
1209 btrfs_set_root_otransid(new_root_item
, trans
->transid
);
1211 old
= btrfs_lock_root_node(root
);
1212 ret
= btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
);
1214 btrfs_tree_unlock(old
);
1215 free_extent_buffer(old
);
1216 btrfs_abort_transaction(trans
, root
, ret
);
1220 btrfs_set_lock_blocking(old
);
1222 ret
= btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
1223 /* clean up in any case */
1224 btrfs_tree_unlock(old
);
1225 free_extent_buffer(old
);
1227 btrfs_abort_transaction(trans
, root
, ret
);
1231 /* see comments in should_cow_block() */
1232 root
->force_cow
= 1;
1235 btrfs_set_root_node(new_root_item
, tmp
);
1236 /* record when the snapshot was created in key.offset */
1237 key
.offset
= trans
->transid
;
1238 ret
= btrfs_insert_root(trans
, tree_root
, &key
, new_root_item
);
1239 btrfs_tree_unlock(tmp
);
1240 free_extent_buffer(tmp
);
1242 btrfs_abort_transaction(trans
, root
, ret
);
1247 * insert root back/forward references
1249 ret
= btrfs_add_root_ref(trans
, tree_root
, objectid
,
1250 parent_root
->root_key
.objectid
,
1251 btrfs_ino(parent_inode
), index
,
1252 dentry
->d_name
.name
, dentry
->d_name
.len
);
1254 btrfs_abort_transaction(trans
, root
, ret
);
1258 key
.offset
= (u64
)-1;
1259 pending
->snap
= btrfs_read_fs_root_no_name(root
->fs_info
, &key
);
1260 if (IS_ERR(pending
->snap
)) {
1261 ret
= PTR_ERR(pending
->snap
);
1262 btrfs_abort_transaction(trans
, root
, ret
);
1266 ret
= btrfs_reloc_post_snapshot(trans
, pending
);
1268 btrfs_abort_transaction(trans
, root
, ret
);
1272 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1274 btrfs_abort_transaction(trans
, root
, ret
);
1278 ret
= btrfs_insert_dir_item(trans
, parent_root
,
1279 dentry
->d_name
.name
, dentry
->d_name
.len
,
1281 BTRFS_FT_DIR
, index
);
1282 /* We have check then name at the beginning, so it is impossible. */
1283 BUG_ON(ret
== -EEXIST
|| ret
== -EOVERFLOW
);
1285 btrfs_abort_transaction(trans
, root
, ret
);
1289 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
1290 dentry
->d_name
.len
* 2);
1291 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
1292 ret
= btrfs_update_inode_fallback(trans
, parent_root
, parent_inode
);
1294 btrfs_abort_transaction(trans
, root
, ret
);
1296 pending
->error
= ret
;
1298 trans
->block_rsv
= rsv
;
1299 trans
->bytes_reserved
= 0;
1301 kfree(new_root_item
);
1302 root_item_alloc_fail
:
1303 btrfs_free_path(path
);
1308 * create all the snapshots we've scheduled for creation
1310 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
,
1311 struct btrfs_fs_info
*fs_info
)
1313 struct btrfs_pending_snapshot
*pending
, *next
;
1314 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
1317 list_for_each_entry_safe(pending
, next
, head
, list
) {
1318 list_del(&pending
->list
);
1319 ret
= create_pending_snapshot(trans
, fs_info
, pending
);
1326 static void update_super_roots(struct btrfs_root
*root
)
1328 struct btrfs_root_item
*root_item
;
1329 struct btrfs_super_block
*super
;
1331 super
= root
->fs_info
->super_copy
;
1333 root_item
= &root
->fs_info
->chunk_root
->root_item
;
1334 super
->chunk_root
= root_item
->bytenr
;
1335 super
->chunk_root_generation
= root_item
->generation
;
1336 super
->chunk_root_level
= root_item
->level
;
1338 root_item
= &root
->fs_info
->tree_root
->root_item
;
1339 super
->root
= root_item
->bytenr
;
1340 super
->generation
= root_item
->generation
;
1341 super
->root_level
= root_item
->level
;
1342 if (btrfs_test_opt(root
, SPACE_CACHE
))
1343 super
->cache_generation
= root_item
->generation
;
1346 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
1349 spin_lock(&info
->trans_lock
);
1350 if (info
->running_transaction
)
1351 ret
= info
->running_transaction
->in_commit
;
1352 spin_unlock(&info
->trans_lock
);
1356 int btrfs_transaction_blocked(struct btrfs_fs_info
*info
)
1359 spin_lock(&info
->trans_lock
);
1360 if (info
->running_transaction
)
1361 ret
= info
->running_transaction
->blocked
;
1362 spin_unlock(&info
->trans_lock
);
1367 * wait for the current transaction commit to start and block subsequent
1370 static void wait_current_trans_commit_start(struct btrfs_root
*root
,
1371 struct btrfs_transaction
*trans
)
1373 wait_event(root
->fs_info
->transaction_blocked_wait
, trans
->in_commit
);
1377 * wait for the current transaction to start and then become unblocked.
1380 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root
*root
,
1381 struct btrfs_transaction
*trans
)
1383 wait_event(root
->fs_info
->transaction_wait
,
1384 trans
->commit_done
|| (trans
->in_commit
&& !trans
->blocked
));
1388 * commit transactions asynchronously. once btrfs_commit_transaction_async
1389 * returns, any subsequent transaction will not be allowed to join.
1391 struct btrfs_async_commit
{
1392 struct btrfs_trans_handle
*newtrans
;
1393 struct btrfs_root
*root
;
1394 struct work_struct work
;
1397 static void do_async_commit(struct work_struct
*work
)
1399 struct btrfs_async_commit
*ac
=
1400 container_of(work
, struct btrfs_async_commit
, work
);
1403 * We've got freeze protection passed with the transaction.
1404 * Tell lockdep about it.
1406 if (ac
->newtrans
->type
< TRANS_JOIN_NOLOCK
)
1408 &ac
->root
->fs_info
->sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
1411 current
->journal_info
= ac
->newtrans
;
1413 btrfs_commit_transaction(ac
->newtrans
, ac
->root
);
1417 int btrfs_commit_transaction_async(struct btrfs_trans_handle
*trans
,
1418 struct btrfs_root
*root
,
1419 int wait_for_unblock
)
1421 struct btrfs_async_commit
*ac
;
1422 struct btrfs_transaction
*cur_trans
;
1424 ac
= kmalloc(sizeof(*ac
), GFP_NOFS
);
1428 INIT_WORK(&ac
->work
, do_async_commit
);
1430 ac
->newtrans
= btrfs_join_transaction(root
);
1431 if (IS_ERR(ac
->newtrans
)) {
1432 int err
= PTR_ERR(ac
->newtrans
);
1437 /* take transaction reference */
1438 cur_trans
= trans
->transaction
;
1439 atomic_inc(&cur_trans
->use_count
);
1441 btrfs_end_transaction(trans
, root
);
1444 * Tell lockdep we've released the freeze rwsem, since the
1445 * async commit thread will be the one to unlock it.
1447 if (trans
->type
< TRANS_JOIN_NOLOCK
)
1449 &root
->fs_info
->sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
1452 schedule_work(&ac
->work
);
1454 /* wait for transaction to start and unblock */
1455 if (wait_for_unblock
)
1456 wait_current_trans_commit_start_and_unblock(root
, cur_trans
);
1458 wait_current_trans_commit_start(root
, cur_trans
);
1460 if (current
->journal_info
== trans
)
1461 current
->journal_info
= NULL
;
1463 put_transaction(cur_trans
);
1468 static void cleanup_transaction(struct btrfs_trans_handle
*trans
,
1469 struct btrfs_root
*root
, int err
)
1471 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1474 WARN_ON(trans
->use_count
> 1);
1476 btrfs_abort_transaction(trans
, root
, err
);
1478 spin_lock(&root
->fs_info
->trans_lock
);
1481 * If the transaction is removed from the list, it means this
1482 * transaction has been committed successfully, so it is impossible
1483 * to call the cleanup function.
1485 BUG_ON(list_empty(&cur_trans
->list
));
1487 list_del_init(&cur_trans
->list
);
1488 if (cur_trans
== root
->fs_info
->running_transaction
) {
1489 root
->fs_info
->trans_no_join
= 1;
1490 spin_unlock(&root
->fs_info
->trans_lock
);
1491 wait_event(cur_trans
->writer_wait
,
1492 atomic_read(&cur_trans
->num_writers
) == 1);
1494 spin_lock(&root
->fs_info
->trans_lock
);
1495 root
->fs_info
->running_transaction
= NULL
;
1497 spin_unlock(&root
->fs_info
->trans_lock
);
1499 btrfs_cleanup_one_transaction(trans
->transaction
, root
);
1501 put_transaction(cur_trans
);
1502 put_transaction(cur_trans
);
1504 trace_btrfs_transaction_commit(root
);
1506 btrfs_scrub_continue(root
);
1508 if (current
->journal_info
== trans
)
1509 current
->journal_info
= NULL
;
1511 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1513 spin_lock(&root
->fs_info
->trans_lock
);
1514 root
->fs_info
->trans_no_join
= 0;
1515 spin_unlock(&root
->fs_info
->trans_lock
);
1518 static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle
*trans
,
1519 struct btrfs_root
*root
)
1521 int flush_on_commit
= btrfs_test_opt(root
, FLUSHONCOMMIT
);
1524 if (flush_on_commit
) {
1525 ret
= btrfs_start_all_delalloc_inodes(root
->fs_info
, 1);
1528 btrfs_wait_all_ordered_extents(root
->fs_info
, 1);
1531 ret
= btrfs_run_delayed_items(trans
, root
);
1536 * running the delayed items may have added new refs. account
1537 * them now so that they hinder processing of more delayed refs
1538 * as little as possible.
1540 btrfs_delayed_refs_qgroup_accounting(trans
, root
->fs_info
);
1543 * rename don't use btrfs_join_transaction, so, once we
1544 * set the transaction to blocked above, we aren't going
1545 * to get any new ordered operations. We can safely run
1546 * it here and no for sure that nothing new will be added
1549 ret
= btrfs_run_ordered_operations(trans
, root
, 1);
1555 * btrfs_transaction state sequence:
1556 * in_commit = 0, blocked = 0 (initial)
1557 * in_commit = 1, blocked = 1
1561 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
,
1562 struct btrfs_root
*root
)
1564 unsigned long joined
= 0;
1565 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1566 struct btrfs_transaction
*prev_trans
= NULL
;
1569 int should_grow
= 0;
1570 unsigned long now
= get_seconds();
1572 ret
= btrfs_run_ordered_operations(trans
, root
, 0);
1574 btrfs_abort_transaction(trans
, root
, ret
);
1575 btrfs_end_transaction(trans
, root
);
1579 /* Stop the commit early if ->aborted is set */
1580 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1581 ret
= cur_trans
->aborted
;
1582 btrfs_end_transaction(trans
, root
);
1586 /* make a pass through all the delayed refs we have so far
1587 * any runnings procs may add more while we are here
1589 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1591 btrfs_end_transaction(trans
, root
);
1595 btrfs_trans_release_metadata(trans
, root
);
1596 trans
->block_rsv
= NULL
;
1597 if (trans
->qgroup_reserved
) {
1598 btrfs_qgroup_free(root
, trans
->qgroup_reserved
);
1599 trans
->qgroup_reserved
= 0;
1602 cur_trans
= trans
->transaction
;
1605 * set the flushing flag so procs in this transaction have to
1606 * start sending their work down.
1608 cur_trans
->delayed_refs
.flushing
= 1;
1610 if (!list_empty(&trans
->new_bgs
))
1611 btrfs_create_pending_block_groups(trans
, root
);
1613 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1615 btrfs_end_transaction(trans
, root
);
1619 spin_lock(&cur_trans
->commit_lock
);
1620 if (cur_trans
->in_commit
) {
1621 spin_unlock(&cur_trans
->commit_lock
);
1622 atomic_inc(&cur_trans
->use_count
);
1623 ret
= btrfs_end_transaction(trans
, root
);
1625 wait_for_commit(root
, cur_trans
);
1627 put_transaction(cur_trans
);
1632 trans
->transaction
->in_commit
= 1;
1633 trans
->transaction
->blocked
= 1;
1634 spin_unlock(&cur_trans
->commit_lock
);
1635 wake_up(&root
->fs_info
->transaction_blocked_wait
);
1637 spin_lock(&root
->fs_info
->trans_lock
);
1638 if (cur_trans
->list
.prev
!= &root
->fs_info
->trans_list
) {
1639 prev_trans
= list_entry(cur_trans
->list
.prev
,
1640 struct btrfs_transaction
, list
);
1641 if (!prev_trans
->commit_done
) {
1642 atomic_inc(&prev_trans
->use_count
);
1643 spin_unlock(&root
->fs_info
->trans_lock
);
1645 wait_for_commit(root
, prev_trans
);
1647 put_transaction(prev_trans
);
1649 spin_unlock(&root
->fs_info
->trans_lock
);
1652 spin_unlock(&root
->fs_info
->trans_lock
);
1655 extwriter_counter_dec(cur_trans
, trans
->type
);
1657 if (!btrfs_test_opt(root
, SSD
) &&
1658 (now
< cur_trans
->start_time
|| now
- cur_trans
->start_time
< 1))
1662 joined
= cur_trans
->num_joined
;
1664 WARN_ON(cur_trans
!= trans
->transaction
);
1666 ret
= btrfs_flush_all_pending_stuffs(trans
, root
);
1668 goto cleanup_transaction
;
1670 prepare_to_wait(&cur_trans
->writer_wait
, &wait
,
1671 TASK_UNINTERRUPTIBLE
);
1673 if (extwriter_counter_read(cur_trans
) > 0)
1675 else if (should_grow
)
1676 schedule_timeout(1);
1678 finish_wait(&cur_trans
->writer_wait
, &wait
);
1679 } while (extwriter_counter_read(cur_trans
) > 0 ||
1680 (should_grow
&& cur_trans
->num_joined
!= joined
));
1682 ret
= btrfs_flush_all_pending_stuffs(trans
, root
);
1684 goto cleanup_transaction
;
1687 * Ok now we need to make sure to block out any other joins while we
1688 * commit the transaction. We could have started a join before setting
1689 * no_join so make sure to wait for num_writers to == 1 again.
1691 spin_lock(&root
->fs_info
->trans_lock
);
1692 root
->fs_info
->trans_no_join
= 1;
1693 spin_unlock(&root
->fs_info
->trans_lock
);
1694 wait_event(cur_trans
->writer_wait
,
1695 atomic_read(&cur_trans
->num_writers
) == 1);
1697 /* ->aborted might be set after the previous check, so check it */
1698 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1699 ret
= cur_trans
->aborted
;
1700 goto cleanup_transaction
;
1703 * the reloc mutex makes sure that we stop
1704 * the balancing code from coming in and moving
1705 * extents around in the middle of the commit
1707 mutex_lock(&root
->fs_info
->reloc_mutex
);
1710 * We needn't worry about the delayed items because we will
1711 * deal with them in create_pending_snapshot(), which is the
1712 * core function of the snapshot creation.
1714 ret
= create_pending_snapshots(trans
, root
->fs_info
);
1716 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1717 goto cleanup_transaction
;
1721 * We insert the dir indexes of the snapshots and update the inode
1722 * of the snapshots' parents after the snapshot creation, so there
1723 * are some delayed items which are not dealt with. Now deal with
1726 * We needn't worry that this operation will corrupt the snapshots,
1727 * because all the tree which are snapshoted will be forced to COW
1728 * the nodes and leaves.
1730 ret
= btrfs_run_delayed_items(trans
, root
);
1732 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1733 goto cleanup_transaction
;
1736 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1738 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1739 goto cleanup_transaction
;
1743 * make sure none of the code above managed to slip in a
1746 btrfs_assert_delayed_root_empty(root
);
1748 WARN_ON(cur_trans
!= trans
->transaction
);
1750 btrfs_scrub_pause(root
);
1751 /* btrfs_commit_tree_roots is responsible for getting the
1752 * various roots consistent with each other. Every pointer
1753 * in the tree of tree roots has to point to the most up to date
1754 * root for every subvolume and other tree. So, we have to keep
1755 * the tree logging code from jumping in and changing any
1758 * At this point in the commit, there can't be any tree-log
1759 * writers, but a little lower down we drop the trans mutex
1760 * and let new people in. By holding the tree_log_mutex
1761 * from now until after the super is written, we avoid races
1762 * with the tree-log code.
1764 mutex_lock(&root
->fs_info
->tree_log_mutex
);
1766 ret
= commit_fs_roots(trans
, root
);
1768 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1769 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1770 goto cleanup_transaction
;
1773 /* commit_fs_roots gets rid of all the tree log roots, it is now
1774 * safe to free the root of tree log roots
1776 btrfs_free_log_root_tree(trans
, root
->fs_info
);
1778 ret
= commit_cowonly_roots(trans
, root
);
1780 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1781 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1782 goto cleanup_transaction
;
1786 * The tasks which save the space cache and inode cache may also
1787 * update ->aborted, check it.
1789 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1790 ret
= cur_trans
->aborted
;
1791 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1792 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1793 goto cleanup_transaction
;
1796 btrfs_prepare_extent_commit(trans
, root
);
1798 cur_trans
= root
->fs_info
->running_transaction
;
1800 btrfs_set_root_node(&root
->fs_info
->tree_root
->root_item
,
1801 root
->fs_info
->tree_root
->node
);
1802 switch_commit_root(root
->fs_info
->tree_root
);
1804 btrfs_set_root_node(&root
->fs_info
->chunk_root
->root_item
,
1805 root
->fs_info
->chunk_root
->node
);
1806 switch_commit_root(root
->fs_info
->chunk_root
);
1808 assert_qgroups_uptodate(trans
);
1809 update_super_roots(root
);
1811 if (!root
->fs_info
->log_root_recovering
) {
1812 btrfs_set_super_log_root(root
->fs_info
->super_copy
, 0);
1813 btrfs_set_super_log_root_level(root
->fs_info
->super_copy
, 0);
1816 memcpy(root
->fs_info
->super_for_commit
, root
->fs_info
->super_copy
,
1817 sizeof(*root
->fs_info
->super_copy
));
1819 trans
->transaction
->blocked
= 0;
1820 spin_lock(&root
->fs_info
->trans_lock
);
1821 root
->fs_info
->running_transaction
= NULL
;
1822 root
->fs_info
->trans_no_join
= 0;
1823 spin_unlock(&root
->fs_info
->trans_lock
);
1824 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1826 wake_up(&root
->fs_info
->transaction_wait
);
1828 ret
= btrfs_write_and_wait_transaction(trans
, root
);
1830 btrfs_error(root
->fs_info
, ret
,
1831 "Error while writing out transaction");
1832 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1833 goto cleanup_transaction
;
1836 ret
= write_ctree_super(trans
, root
, 0);
1838 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1839 goto cleanup_transaction
;
1843 * the super is written, we can safely allow the tree-loggers
1844 * to go about their business
1846 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1848 btrfs_finish_extent_commit(trans
, root
);
1850 cur_trans
->commit_done
= 1;
1852 root
->fs_info
->last_trans_committed
= cur_trans
->transid
;
1854 wake_up(&cur_trans
->commit_wait
);
1856 spin_lock(&root
->fs_info
->trans_lock
);
1857 list_del_init(&cur_trans
->list
);
1858 spin_unlock(&root
->fs_info
->trans_lock
);
1860 put_transaction(cur_trans
);
1861 put_transaction(cur_trans
);
1863 if (trans
->type
& __TRANS_FREEZABLE
)
1864 sb_end_intwrite(root
->fs_info
->sb
);
1866 trace_btrfs_transaction_commit(root
);
1868 btrfs_scrub_continue(root
);
1870 if (current
->journal_info
== trans
)
1871 current
->journal_info
= NULL
;
1873 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1875 if (current
!= root
->fs_info
->transaction_kthread
)
1876 btrfs_run_delayed_iputs(root
);
1880 cleanup_transaction
:
1881 btrfs_trans_release_metadata(trans
, root
);
1882 trans
->block_rsv
= NULL
;
1883 if (trans
->qgroup_reserved
) {
1884 btrfs_qgroup_free(root
, trans
->qgroup_reserved
);
1885 trans
->qgroup_reserved
= 0;
1887 btrfs_warn(root
->fs_info
, "Skipping commit of aborted transaction.");
1888 if (current
->journal_info
== trans
)
1889 current
->journal_info
= NULL
;
1890 cleanup_transaction(trans
, root
, ret
);
1896 * return < 0 if error
1897 * 0 if there are no more dead_roots at the time of call
1898 * 1 there are more to be processed, call me again
1900 * The return value indicates there are certainly more snapshots to delete, but
1901 * if there comes a new one during processing, it may return 0. We don't mind,
1902 * because btrfs_commit_super will poke cleaner thread and it will process it a
1903 * few seconds later.
1905 int btrfs_clean_one_deleted_snapshot(struct btrfs_root
*root
)
1908 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1910 spin_lock(&fs_info
->trans_lock
);
1911 if (list_empty(&fs_info
->dead_roots
)) {
1912 spin_unlock(&fs_info
->trans_lock
);
1915 root
= list_first_entry(&fs_info
->dead_roots
,
1916 struct btrfs_root
, root_list
);
1917 list_del(&root
->root_list
);
1918 spin_unlock(&fs_info
->trans_lock
);
1920 pr_debug("btrfs: cleaner removing %llu\n",
1921 (unsigned long long)root
->objectid
);
1923 btrfs_kill_all_delayed_nodes(root
);
1925 if (btrfs_header_backref_rev(root
->node
) <
1926 BTRFS_MIXED_BACKREF_REV
)
1927 ret
= btrfs_drop_snapshot(root
, NULL
, 0, 0);
1929 ret
= btrfs_drop_snapshot(root
, NULL
, 1, 0);
1931 * If we encounter a transaction abort during snapshot cleaning, we
1932 * don't want to crash here
1934 BUG_ON(ret
< 0 && ret
!= -EAGAIN
&& ret
!= -EROFS
);