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 unsigned int btrfs_blocked_trans_types
[TRANS_STATE_MAX
] = {
38 [TRANS_STATE_RUNNING
] = 0U,
39 [TRANS_STATE_BLOCKED
] = (__TRANS_USERSPACE
|
41 [TRANS_STATE_COMMIT_START
] = (__TRANS_USERSPACE
|
44 [TRANS_STATE_COMMIT_DOING
] = (__TRANS_USERSPACE
|
48 [TRANS_STATE_UNBLOCKED
] = (__TRANS_USERSPACE
|
53 [TRANS_STATE_COMPLETED
] = (__TRANS_USERSPACE
|
60 void btrfs_put_transaction(struct btrfs_transaction
*transaction
)
62 WARN_ON(atomic_read(&transaction
->use_count
) == 0);
63 if (atomic_dec_and_test(&transaction
->use_count
)) {
64 BUG_ON(!list_empty(&transaction
->list
));
65 WARN_ON(transaction
->delayed_refs
.root
.rb_node
);
66 while (!list_empty(&transaction
->pending_chunks
)) {
67 struct extent_map
*em
;
69 em
= list_first_entry(&transaction
->pending_chunks
,
70 struct extent_map
, list
);
71 list_del_init(&em
->list
);
74 kmem_cache_free(btrfs_transaction_cachep
, transaction
);
78 static noinline
void switch_commit_root(struct btrfs_root
*root
)
80 free_extent_buffer(root
->commit_root
);
81 root
->commit_root
= btrfs_root_node(root
);
84 static inline void extwriter_counter_inc(struct btrfs_transaction
*trans
,
87 if (type
& TRANS_EXTWRITERS
)
88 atomic_inc(&trans
->num_extwriters
);
91 static inline void extwriter_counter_dec(struct btrfs_transaction
*trans
,
94 if (type
& TRANS_EXTWRITERS
)
95 atomic_dec(&trans
->num_extwriters
);
98 static inline void extwriter_counter_init(struct btrfs_transaction
*trans
,
101 atomic_set(&trans
->num_extwriters
, ((type
& TRANS_EXTWRITERS
) ? 1 : 0));
104 static inline int extwriter_counter_read(struct btrfs_transaction
*trans
)
106 return atomic_read(&trans
->num_extwriters
);
110 * either allocate a new transaction or hop into the existing one
112 static noinline
int join_transaction(struct btrfs_root
*root
, unsigned int type
)
114 struct btrfs_transaction
*cur_trans
;
115 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
117 spin_lock(&fs_info
->trans_lock
);
119 /* The file system has been taken offline. No new transactions. */
120 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
121 spin_unlock(&fs_info
->trans_lock
);
125 cur_trans
= fs_info
->running_transaction
;
127 if (cur_trans
->aborted
) {
128 spin_unlock(&fs_info
->trans_lock
);
129 return cur_trans
->aborted
;
131 if (btrfs_blocked_trans_types
[cur_trans
->state
] & type
) {
132 spin_unlock(&fs_info
->trans_lock
);
135 atomic_inc(&cur_trans
->use_count
);
136 atomic_inc(&cur_trans
->num_writers
);
137 extwriter_counter_inc(cur_trans
, type
);
138 spin_unlock(&fs_info
->trans_lock
);
141 spin_unlock(&fs_info
->trans_lock
);
144 * If we are ATTACH, we just want to catch the current transaction,
145 * and commit it. If there is no transaction, just return ENOENT.
147 if (type
== TRANS_ATTACH
)
151 * JOIN_NOLOCK only happens during the transaction commit, so
152 * it is impossible that ->running_transaction is NULL
154 BUG_ON(type
== TRANS_JOIN_NOLOCK
);
156 cur_trans
= kmem_cache_alloc(btrfs_transaction_cachep
, GFP_NOFS
);
160 spin_lock(&fs_info
->trans_lock
);
161 if (fs_info
->running_transaction
) {
163 * someone started a transaction after we unlocked. Make sure
164 * to redo the checks above
166 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
168 } else if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
169 spin_unlock(&fs_info
->trans_lock
);
170 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
174 atomic_set(&cur_trans
->num_writers
, 1);
175 extwriter_counter_init(cur_trans
, type
);
176 init_waitqueue_head(&cur_trans
->writer_wait
);
177 init_waitqueue_head(&cur_trans
->commit_wait
);
178 cur_trans
->state
= TRANS_STATE_RUNNING
;
180 * One for this trans handle, one so it will live on until we
181 * commit the transaction.
183 atomic_set(&cur_trans
->use_count
, 2);
184 cur_trans
->start_time
= get_seconds();
186 cur_trans
->delayed_refs
.root
= RB_ROOT
;
187 cur_trans
->delayed_refs
.num_entries
= 0;
188 cur_trans
->delayed_refs
.num_heads_ready
= 0;
189 cur_trans
->delayed_refs
.num_heads
= 0;
190 cur_trans
->delayed_refs
.flushing
= 0;
191 cur_trans
->delayed_refs
.run_delayed_start
= 0;
194 * although the tree mod log is per file system and not per transaction,
195 * the log must never go across transaction boundaries.
198 if (!list_empty(&fs_info
->tree_mod_seq_list
))
199 WARN(1, KERN_ERR
"btrfs: tree_mod_seq_list not empty when "
200 "creating a fresh transaction\n");
201 if (!RB_EMPTY_ROOT(&fs_info
->tree_mod_log
))
202 WARN(1, KERN_ERR
"btrfs: tree_mod_log rb tree not empty when "
203 "creating a fresh transaction\n");
204 atomic64_set(&fs_info
->tree_mod_seq
, 0);
206 spin_lock_init(&cur_trans
->delayed_refs
.lock
);
207 atomic_set(&cur_trans
->delayed_refs
.procs_running_refs
, 0);
208 atomic_set(&cur_trans
->delayed_refs
.ref_seq
, 0);
209 init_waitqueue_head(&cur_trans
->delayed_refs
.wait
);
211 INIT_LIST_HEAD(&cur_trans
->pending_snapshots
);
212 INIT_LIST_HEAD(&cur_trans
->ordered_operations
);
213 INIT_LIST_HEAD(&cur_trans
->pending_chunks
);
214 list_add_tail(&cur_trans
->list
, &fs_info
->trans_list
);
215 extent_io_tree_init(&cur_trans
->dirty_pages
,
216 fs_info
->btree_inode
->i_mapping
);
217 fs_info
->generation
++;
218 cur_trans
->transid
= fs_info
->generation
;
219 fs_info
->running_transaction
= cur_trans
;
220 cur_trans
->aborted
= 0;
221 spin_unlock(&fs_info
->trans_lock
);
227 * this does all the record keeping required to make sure that a reference
228 * counted root is properly recorded in a given transaction. This is required
229 * to make sure the old root from before we joined the transaction is deleted
230 * when the transaction commits
232 static int record_root_in_trans(struct btrfs_trans_handle
*trans
,
233 struct btrfs_root
*root
)
235 if (root
->ref_cows
&& root
->last_trans
< trans
->transid
) {
236 WARN_ON(root
== root
->fs_info
->extent_root
);
237 WARN_ON(root
->commit_root
!= root
->node
);
240 * see below for in_trans_setup usage rules
241 * we have the reloc mutex held now, so there
242 * is only one writer in this function
244 root
->in_trans_setup
= 1;
246 /* make sure readers find in_trans_setup before
247 * they find our root->last_trans update
251 spin_lock(&root
->fs_info
->fs_roots_radix_lock
);
252 if (root
->last_trans
== trans
->transid
) {
253 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
256 radix_tree_tag_set(&root
->fs_info
->fs_roots_radix
,
257 (unsigned long)root
->root_key
.objectid
,
258 BTRFS_ROOT_TRANS_TAG
);
259 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
260 root
->last_trans
= trans
->transid
;
262 /* this is pretty tricky. We don't want to
263 * take the relocation lock in btrfs_record_root_in_trans
264 * unless we're really doing the first setup for this root in
267 * Normally we'd use root->last_trans as a flag to decide
268 * if we want to take the expensive mutex.
270 * But, we have to set root->last_trans before we
271 * init the relocation root, otherwise, we trip over warnings
272 * in ctree.c. The solution used here is to flag ourselves
273 * with root->in_trans_setup. When this is 1, we're still
274 * fixing up the reloc trees and everyone must wait.
276 * When this is zero, they can trust root->last_trans and fly
277 * through btrfs_record_root_in_trans without having to take the
278 * lock. smp_wmb() makes sure that all the writes above are
279 * done before we pop in the zero below
281 btrfs_init_reloc_root(trans
, root
);
283 root
->in_trans_setup
= 0;
289 int btrfs_record_root_in_trans(struct btrfs_trans_handle
*trans
,
290 struct btrfs_root
*root
)
296 * see record_root_in_trans for comments about in_trans_setup usage
300 if (root
->last_trans
== trans
->transid
&&
301 !root
->in_trans_setup
)
304 mutex_lock(&root
->fs_info
->reloc_mutex
);
305 record_root_in_trans(trans
, root
);
306 mutex_unlock(&root
->fs_info
->reloc_mutex
);
311 static inline int is_transaction_blocked(struct btrfs_transaction
*trans
)
313 return (trans
->state
>= TRANS_STATE_BLOCKED
&&
314 trans
->state
< TRANS_STATE_UNBLOCKED
&&
318 /* wait for commit against the current transaction to become unblocked
319 * when this is done, it is safe to start a new transaction, but the current
320 * transaction might not be fully on disk.
322 static void wait_current_trans(struct btrfs_root
*root
)
324 struct btrfs_transaction
*cur_trans
;
326 spin_lock(&root
->fs_info
->trans_lock
);
327 cur_trans
= root
->fs_info
->running_transaction
;
328 if (cur_trans
&& is_transaction_blocked(cur_trans
)) {
329 atomic_inc(&cur_trans
->use_count
);
330 spin_unlock(&root
->fs_info
->trans_lock
);
332 wait_event(root
->fs_info
->transaction_wait
,
333 cur_trans
->state
>= TRANS_STATE_UNBLOCKED
||
335 btrfs_put_transaction(cur_trans
);
337 spin_unlock(&root
->fs_info
->trans_lock
);
341 static int may_wait_transaction(struct btrfs_root
*root
, int type
)
343 if (root
->fs_info
->log_root_recovering
)
346 if (type
== TRANS_USERSPACE
)
349 if (type
== TRANS_START
&&
350 !atomic_read(&root
->fs_info
->open_ioctl_trans
))
356 static inline bool need_reserve_reloc_root(struct btrfs_root
*root
)
358 if (!root
->fs_info
->reloc_ctl
||
360 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
367 static struct btrfs_trans_handle
*
368 start_transaction(struct btrfs_root
*root
, u64 num_items
, unsigned int type
,
369 enum btrfs_reserve_flush_enum flush
)
371 struct btrfs_trans_handle
*h
;
372 struct btrfs_transaction
*cur_trans
;
374 u64 qgroup_reserved
= 0;
375 bool reloc_reserved
= false;
378 if (test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
379 return ERR_PTR(-EROFS
);
381 if (current
->journal_info
) {
382 WARN_ON(type
& TRANS_EXTWRITERS
);
383 h
= current
->journal_info
;
385 WARN_ON(h
->use_count
> 2);
386 h
->orig_rsv
= h
->block_rsv
;
392 * Do the reservation before we join the transaction so we can do all
393 * the appropriate flushing if need be.
395 if (num_items
> 0 && root
!= root
->fs_info
->chunk_root
) {
396 if (root
->fs_info
->quota_enabled
&&
397 is_fstree(root
->root_key
.objectid
)) {
398 qgroup_reserved
= num_items
* root
->leafsize
;
399 ret
= btrfs_qgroup_reserve(root
, qgroup_reserved
);
404 num_bytes
= btrfs_calc_trans_metadata_size(root
, num_items
);
406 * Do the reservation for the relocation root creation
408 if (unlikely(need_reserve_reloc_root(root
))) {
409 num_bytes
+= root
->nodesize
;
410 reloc_reserved
= true;
413 ret
= btrfs_block_rsv_add(root
,
414 &root
->fs_info
->trans_block_rsv
,
420 h
= kmem_cache_alloc(btrfs_trans_handle_cachep
, GFP_NOFS
);
427 * If we are JOIN_NOLOCK we're already committing a transaction and
428 * waiting on this guy, so we don't need to do the sb_start_intwrite
429 * because we're already holding a ref. We need this because we could
430 * have raced in and did an fsync() on a file which can kick a commit
431 * and then we deadlock with somebody doing a freeze.
433 * If we are ATTACH, it means we just want to catch the current
434 * transaction and commit it, so we needn't do sb_start_intwrite().
436 if (type
& __TRANS_FREEZABLE
)
437 sb_start_intwrite(root
->fs_info
->sb
);
439 if (may_wait_transaction(root
, type
))
440 wait_current_trans(root
);
443 ret
= join_transaction(root
, type
);
445 wait_current_trans(root
);
446 if (unlikely(type
== TRANS_ATTACH
))
449 } while (ret
== -EBUSY
);
452 /* We must get the transaction if we are JOIN_NOLOCK. */
453 BUG_ON(type
== TRANS_JOIN_NOLOCK
);
457 cur_trans
= root
->fs_info
->running_transaction
;
459 h
->transid
= cur_trans
->transid
;
460 h
->transaction
= cur_trans
;
462 h
->bytes_reserved
= 0;
464 h
->delayed_ref_updates
= 0;
470 h
->qgroup_reserved
= 0;
471 h
->delayed_ref_elem
.seq
= 0;
473 h
->allocating_chunk
= false;
474 h
->reloc_reserved
= false;
475 INIT_LIST_HEAD(&h
->qgroup_ref_list
);
476 INIT_LIST_HEAD(&h
->new_bgs
);
479 if (cur_trans
->state
>= TRANS_STATE_BLOCKED
&&
480 may_wait_transaction(root
, type
)) {
481 btrfs_commit_transaction(h
, root
);
486 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
487 h
->transid
, num_bytes
, 1);
488 h
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
489 h
->bytes_reserved
= num_bytes
;
490 h
->reloc_reserved
= reloc_reserved
;
492 h
->qgroup_reserved
= qgroup_reserved
;
495 btrfs_record_root_in_trans(h
, root
);
497 if (!current
->journal_info
&& type
!= TRANS_USERSPACE
)
498 current
->journal_info
= h
;
502 if (type
& __TRANS_FREEZABLE
)
503 sb_end_intwrite(root
->fs_info
->sb
);
504 kmem_cache_free(btrfs_trans_handle_cachep
, h
);
507 btrfs_block_rsv_release(root
, &root
->fs_info
->trans_block_rsv
,
511 btrfs_qgroup_free(root
, qgroup_reserved
);
515 struct btrfs_trans_handle
*btrfs_start_transaction(struct btrfs_root
*root
,
518 return start_transaction(root
, num_items
, TRANS_START
,
519 BTRFS_RESERVE_FLUSH_ALL
);
522 struct btrfs_trans_handle
*btrfs_start_transaction_lflush(
523 struct btrfs_root
*root
, int num_items
)
525 return start_transaction(root
, num_items
, TRANS_START
,
526 BTRFS_RESERVE_FLUSH_LIMIT
);
529 struct btrfs_trans_handle
*btrfs_join_transaction(struct btrfs_root
*root
)
531 return start_transaction(root
, 0, TRANS_JOIN
, 0);
534 struct btrfs_trans_handle
*btrfs_join_transaction_nolock(struct btrfs_root
*root
)
536 return start_transaction(root
, 0, TRANS_JOIN_NOLOCK
, 0);
539 struct btrfs_trans_handle
*btrfs_start_ioctl_transaction(struct btrfs_root
*root
)
541 return start_transaction(root
, 0, TRANS_USERSPACE
, 0);
545 * btrfs_attach_transaction() - catch the running transaction
547 * It is used when we want to commit the current the transaction, but
548 * don't want to start a new one.
550 * Note: If this function return -ENOENT, it just means there is no
551 * running transaction. But it is possible that the inactive transaction
552 * is still in the memory, not fully on disk. If you hope there is no
553 * inactive transaction in the fs when -ENOENT is returned, you should
555 * btrfs_attach_transaction_barrier()
557 struct btrfs_trans_handle
*btrfs_attach_transaction(struct btrfs_root
*root
)
559 return start_transaction(root
, 0, TRANS_ATTACH
, 0);
563 * btrfs_attach_transaction_barrier() - catch the running transaction
565 * It is similar to the above function, the differentia is this one
566 * will wait for all the inactive transactions until they fully
569 struct btrfs_trans_handle
*
570 btrfs_attach_transaction_barrier(struct btrfs_root
*root
)
572 struct btrfs_trans_handle
*trans
;
574 trans
= start_transaction(root
, 0, TRANS_ATTACH
, 0);
575 if (IS_ERR(trans
) && PTR_ERR(trans
) == -ENOENT
)
576 btrfs_wait_for_commit(root
, 0);
581 /* wait for a transaction commit to be fully complete */
582 static noinline
void wait_for_commit(struct btrfs_root
*root
,
583 struct btrfs_transaction
*commit
)
585 wait_event(commit
->commit_wait
, commit
->state
== TRANS_STATE_COMPLETED
);
588 int btrfs_wait_for_commit(struct btrfs_root
*root
, u64 transid
)
590 struct btrfs_transaction
*cur_trans
= NULL
, *t
;
594 if (transid
<= root
->fs_info
->last_trans_committed
)
598 /* find specified transaction */
599 spin_lock(&root
->fs_info
->trans_lock
);
600 list_for_each_entry(t
, &root
->fs_info
->trans_list
, list
) {
601 if (t
->transid
== transid
) {
603 atomic_inc(&cur_trans
->use_count
);
607 if (t
->transid
> transid
) {
612 spin_unlock(&root
->fs_info
->trans_lock
);
613 /* The specified transaction doesn't exist */
617 /* find newest transaction that is committing | committed */
618 spin_lock(&root
->fs_info
->trans_lock
);
619 list_for_each_entry_reverse(t
, &root
->fs_info
->trans_list
,
621 if (t
->state
>= TRANS_STATE_COMMIT_START
) {
622 if (t
->state
== TRANS_STATE_COMPLETED
)
625 atomic_inc(&cur_trans
->use_count
);
629 spin_unlock(&root
->fs_info
->trans_lock
);
631 goto out
; /* nothing committing|committed */
634 wait_for_commit(root
, cur_trans
);
635 btrfs_put_transaction(cur_trans
);
640 void btrfs_throttle(struct btrfs_root
*root
)
642 if (!atomic_read(&root
->fs_info
->open_ioctl_trans
))
643 wait_current_trans(root
);
646 static int should_end_transaction(struct btrfs_trans_handle
*trans
,
647 struct btrfs_root
*root
)
649 if (root
->fs_info
->global_block_rsv
.space_info
->full
&&
650 btrfs_should_throttle_delayed_refs(trans
, root
))
653 return !!btrfs_block_rsv_check(root
, &root
->fs_info
->global_block_rsv
, 5);
656 int btrfs_should_end_transaction(struct btrfs_trans_handle
*trans
,
657 struct btrfs_root
*root
)
659 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
664 if (cur_trans
->state
>= TRANS_STATE_BLOCKED
||
665 cur_trans
->delayed_refs
.flushing
)
668 updates
= trans
->delayed_ref_updates
;
669 trans
->delayed_ref_updates
= 0;
671 err
= btrfs_run_delayed_refs(trans
, root
, updates
);
672 if (err
) /* Error code will also eval true */
676 return should_end_transaction(trans
, root
);
679 static int __btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
680 struct btrfs_root
*root
, int throttle
)
682 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
683 struct btrfs_fs_info
*info
= root
->fs_info
;
684 unsigned long cur
= trans
->delayed_ref_updates
;
685 int lock
= (trans
->type
!= TRANS_JOIN_NOLOCK
);
688 if (--trans
->use_count
) {
689 trans
->block_rsv
= trans
->orig_rsv
;
694 * do the qgroup accounting as early as possible
696 err
= btrfs_delayed_refs_qgroup_accounting(trans
, info
);
698 btrfs_trans_release_metadata(trans
, root
);
699 trans
->block_rsv
= NULL
;
701 if (trans
->qgroup_reserved
) {
703 * the same root has to be passed here between start_transaction
704 * and end_transaction. Subvolume quota depends on this.
706 btrfs_qgroup_free(trans
->root
, trans
->qgroup_reserved
);
707 trans
->qgroup_reserved
= 0;
710 if (!list_empty(&trans
->new_bgs
))
711 btrfs_create_pending_block_groups(trans
, root
);
713 trans
->delayed_ref_updates
= 0;
714 if (btrfs_should_throttle_delayed_refs(trans
, root
)) {
715 cur
= max_t(unsigned long, cur
, 1);
716 trans
->delayed_ref_updates
= 0;
717 btrfs_run_delayed_refs(trans
, root
, cur
);
720 btrfs_trans_release_metadata(trans
, root
);
721 trans
->block_rsv
= NULL
;
723 if (!list_empty(&trans
->new_bgs
))
724 btrfs_create_pending_block_groups(trans
, root
);
726 if (lock
&& !atomic_read(&root
->fs_info
->open_ioctl_trans
) &&
727 should_end_transaction(trans
, root
) &&
728 ACCESS_ONCE(cur_trans
->state
) == TRANS_STATE_RUNNING
) {
729 spin_lock(&info
->trans_lock
);
730 if (cur_trans
->state
== TRANS_STATE_RUNNING
)
731 cur_trans
->state
= TRANS_STATE_BLOCKED
;
732 spin_unlock(&info
->trans_lock
);
735 if (lock
&& ACCESS_ONCE(cur_trans
->state
) == TRANS_STATE_BLOCKED
) {
738 * We may race with somebody else here so end up having
739 * to call end_transaction on ourselves again, so inc
743 return btrfs_commit_transaction(trans
, root
);
745 wake_up_process(info
->transaction_kthread
);
749 if (trans
->type
& __TRANS_FREEZABLE
)
750 sb_end_intwrite(root
->fs_info
->sb
);
752 WARN_ON(cur_trans
!= info
->running_transaction
);
753 WARN_ON(atomic_read(&cur_trans
->num_writers
) < 1);
754 atomic_dec(&cur_trans
->num_writers
);
755 extwriter_counter_dec(cur_trans
, trans
->type
);
758 if (waitqueue_active(&cur_trans
->writer_wait
))
759 wake_up(&cur_trans
->writer_wait
);
760 btrfs_put_transaction(cur_trans
);
762 if (current
->journal_info
== trans
)
763 current
->journal_info
= NULL
;
766 btrfs_run_delayed_iputs(root
);
768 if (trans
->aborted
||
769 test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
)) {
770 wake_up_process(info
->transaction_kthread
);
773 assert_qgroups_uptodate(trans
);
775 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
779 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
780 struct btrfs_root
*root
)
782 return __btrfs_end_transaction(trans
, root
, 0);
785 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
,
786 struct btrfs_root
*root
)
788 return __btrfs_end_transaction(trans
, root
, 1);
791 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle
*trans
,
792 struct btrfs_root
*root
)
794 return __btrfs_end_transaction(trans
, root
, 1);
798 * when btree blocks are allocated, they have some corresponding bits set for
799 * them in one of two extent_io trees. This is used to make sure all of
800 * those extents are sent to disk but does not wait on them
802 int btrfs_write_marked_extents(struct btrfs_root
*root
,
803 struct extent_io_tree
*dirty_pages
, int mark
)
807 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
808 struct extent_state
*cached_state
= NULL
;
812 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
813 mark
, &cached_state
)) {
814 convert_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
,
815 mark
, &cached_state
, GFP_NOFS
);
817 err
= filemap_fdatawrite_range(mapping
, start
, end
);
829 * when btree blocks are allocated, they have some corresponding bits set for
830 * them in one of two extent_io trees. This is used to make sure all of
831 * those extents are on disk for transaction or log commit. We wait
832 * on all the pages and clear them from the dirty pages state tree
834 int btrfs_wait_marked_extents(struct btrfs_root
*root
,
835 struct extent_io_tree
*dirty_pages
, int mark
)
839 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
840 struct extent_state
*cached_state
= NULL
;
844 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
845 EXTENT_NEED_WAIT
, &cached_state
)) {
846 clear_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
,
847 0, 0, &cached_state
, GFP_NOFS
);
848 err
= filemap_fdatawait_range(mapping
, start
, end
);
860 * when btree blocks are allocated, they have some corresponding bits set for
861 * them in one of two extent_io trees. This is used to make sure all of
862 * those extents are on disk for transaction or log commit
864 static int btrfs_write_and_wait_marked_extents(struct btrfs_root
*root
,
865 struct extent_io_tree
*dirty_pages
, int mark
)
869 struct blk_plug plug
;
871 blk_start_plug(&plug
);
872 ret
= btrfs_write_marked_extents(root
, dirty_pages
, mark
);
873 blk_finish_plug(&plug
);
874 ret2
= btrfs_wait_marked_extents(root
, dirty_pages
, mark
);
883 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
,
884 struct btrfs_root
*root
)
886 if (!trans
|| !trans
->transaction
) {
887 struct inode
*btree_inode
;
888 btree_inode
= root
->fs_info
->btree_inode
;
889 return filemap_write_and_wait(btree_inode
->i_mapping
);
891 return btrfs_write_and_wait_marked_extents(root
,
892 &trans
->transaction
->dirty_pages
,
897 * this is used to update the root pointer in the tree of tree roots.
899 * But, in the case of the extent allocation tree, updating the root
900 * pointer may allocate blocks which may change the root of the extent
903 * So, this loops and repeats and makes sure the cowonly root didn't
904 * change while the root pointer was being updated in the metadata.
906 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
907 struct btrfs_root
*root
)
912 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
914 old_root_used
= btrfs_root_used(&root
->root_item
);
915 btrfs_write_dirty_block_groups(trans
, root
);
918 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
919 if (old_root_bytenr
== root
->node
->start
&&
920 old_root_used
== btrfs_root_used(&root
->root_item
))
923 btrfs_set_root_node(&root
->root_item
, root
->node
);
924 ret
= btrfs_update_root(trans
, tree_root
,
930 old_root_used
= btrfs_root_used(&root
->root_item
);
931 ret
= btrfs_write_dirty_block_groups(trans
, root
);
936 if (root
!= root
->fs_info
->extent_root
)
937 switch_commit_root(root
);
943 * update all the cowonly tree roots on disk
945 * The error handling in this function may not be obvious. Any of the
946 * failures will cause the file system to go offline. We still need
947 * to clean up the delayed refs.
949 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
,
950 struct btrfs_root
*root
)
952 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
953 struct list_head
*next
;
954 struct extent_buffer
*eb
;
957 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
961 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
962 ret
= btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
,
964 btrfs_tree_unlock(eb
);
965 free_extent_buffer(eb
);
970 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
974 ret
= btrfs_run_dev_stats(trans
, root
->fs_info
);
977 ret
= btrfs_run_dev_replace(trans
, root
->fs_info
);
980 ret
= btrfs_run_qgroups(trans
, root
->fs_info
);
984 /* run_qgroups might have added some more refs */
985 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
989 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
990 next
= fs_info
->dirty_cowonly_roots
.next
;
992 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
994 ret
= update_cowonly_root(trans
, root
);
999 down_write(&fs_info
->extent_commit_sem
);
1000 switch_commit_root(fs_info
->extent_root
);
1001 up_write(&fs_info
->extent_commit_sem
);
1003 btrfs_after_dev_replace_commit(fs_info
);
1009 * dead roots are old snapshots that need to be deleted. This allocates
1010 * a dirty root struct and adds it into the list of dead roots that need to
1013 void btrfs_add_dead_root(struct btrfs_root
*root
)
1015 spin_lock(&root
->fs_info
->trans_lock
);
1016 if (list_empty(&root
->root_list
))
1017 list_add_tail(&root
->root_list
, &root
->fs_info
->dead_roots
);
1018 spin_unlock(&root
->fs_info
->trans_lock
);
1022 * update all the cowonly tree roots on disk
1024 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
,
1025 struct btrfs_root
*root
)
1027 struct btrfs_root
*gang
[8];
1028 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1033 spin_lock(&fs_info
->fs_roots_radix_lock
);
1035 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
1038 BTRFS_ROOT_TRANS_TAG
);
1041 for (i
= 0; i
< ret
; i
++) {
1043 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
1044 (unsigned long)root
->root_key
.objectid
,
1045 BTRFS_ROOT_TRANS_TAG
);
1046 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1048 btrfs_free_log(trans
, root
);
1049 btrfs_update_reloc_root(trans
, root
);
1050 btrfs_orphan_commit_root(trans
, root
);
1052 btrfs_save_ino_cache(root
, trans
);
1054 /* see comments in should_cow_block() */
1055 root
->force_cow
= 0;
1058 if (root
->commit_root
!= root
->node
) {
1059 mutex_lock(&root
->fs_commit_mutex
);
1060 switch_commit_root(root
);
1061 btrfs_unpin_free_ino(root
);
1062 mutex_unlock(&root
->fs_commit_mutex
);
1064 btrfs_set_root_node(&root
->root_item
,
1068 err
= btrfs_update_root(trans
, fs_info
->tree_root
,
1071 spin_lock(&fs_info
->fs_roots_radix_lock
);
1076 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1081 * defrag a given btree.
1082 * Every leaf in the btree is read and defragged.
1084 int btrfs_defrag_root(struct btrfs_root
*root
)
1086 struct btrfs_fs_info
*info
= root
->fs_info
;
1087 struct btrfs_trans_handle
*trans
;
1090 if (xchg(&root
->defrag_running
, 1))
1094 trans
= btrfs_start_transaction(root
, 0);
1096 return PTR_ERR(trans
);
1098 ret
= btrfs_defrag_leaves(trans
, root
);
1100 btrfs_end_transaction(trans
, root
);
1101 btrfs_btree_balance_dirty(info
->tree_root
);
1104 if (btrfs_fs_closing(root
->fs_info
) || ret
!= -EAGAIN
)
1107 if (btrfs_defrag_cancelled(root
->fs_info
)) {
1108 printk(KERN_DEBUG
"btrfs: defrag_root cancelled\n");
1113 root
->defrag_running
= 0;
1118 * new snapshots need to be created at a very specific time in the
1119 * transaction commit. This does the actual creation.
1122 * If the error which may affect the commitment of the current transaction
1123 * happens, we should return the error number. If the error which just affect
1124 * the creation of the pending snapshots, just return 0.
1126 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
1127 struct btrfs_fs_info
*fs_info
,
1128 struct btrfs_pending_snapshot
*pending
)
1130 struct btrfs_key key
;
1131 struct btrfs_root_item
*new_root_item
;
1132 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1133 struct btrfs_root
*root
= pending
->root
;
1134 struct btrfs_root
*parent_root
;
1135 struct btrfs_block_rsv
*rsv
;
1136 struct inode
*parent_inode
;
1137 struct btrfs_path
*path
;
1138 struct btrfs_dir_item
*dir_item
;
1139 struct dentry
*dentry
;
1140 struct extent_buffer
*tmp
;
1141 struct extent_buffer
*old
;
1142 struct timespec cur_time
= CURRENT_TIME
;
1150 path
= btrfs_alloc_path();
1152 pending
->error
= -ENOMEM
;
1156 new_root_item
= kmalloc(sizeof(*new_root_item
), GFP_NOFS
);
1157 if (!new_root_item
) {
1158 pending
->error
= -ENOMEM
;
1159 goto root_item_alloc_fail
;
1162 pending
->error
= btrfs_find_free_objectid(tree_root
, &objectid
);
1164 goto no_free_objectid
;
1166 btrfs_reloc_pre_snapshot(trans
, pending
, &to_reserve
);
1168 if (to_reserve
> 0) {
1169 pending
->error
= btrfs_block_rsv_add(root
,
1170 &pending
->block_rsv
,
1172 BTRFS_RESERVE_NO_FLUSH
);
1174 goto no_free_objectid
;
1177 pending
->error
= btrfs_qgroup_inherit(trans
, fs_info
,
1178 root
->root_key
.objectid
,
1179 objectid
, pending
->inherit
);
1181 goto no_free_objectid
;
1183 key
.objectid
= objectid
;
1184 key
.offset
= (u64
)-1;
1185 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1187 rsv
= trans
->block_rsv
;
1188 trans
->block_rsv
= &pending
->block_rsv
;
1189 trans
->bytes_reserved
= trans
->block_rsv
->reserved
;
1191 dentry
= pending
->dentry
;
1192 parent_inode
= pending
->dir
;
1193 parent_root
= BTRFS_I(parent_inode
)->root
;
1194 record_root_in_trans(trans
, parent_root
);
1197 * insert the directory item
1199 ret
= btrfs_set_inode_index(parent_inode
, &index
);
1200 BUG_ON(ret
); /* -ENOMEM */
1202 /* check if there is a file/dir which has the same name. */
1203 dir_item
= btrfs_lookup_dir_item(NULL
, parent_root
, path
,
1204 btrfs_ino(parent_inode
),
1205 dentry
->d_name
.name
,
1206 dentry
->d_name
.len
, 0);
1207 if (dir_item
!= NULL
&& !IS_ERR(dir_item
)) {
1208 pending
->error
= -EEXIST
;
1209 goto dir_item_existed
;
1210 } else if (IS_ERR(dir_item
)) {
1211 ret
= PTR_ERR(dir_item
);
1212 btrfs_abort_transaction(trans
, root
, ret
);
1215 btrfs_release_path(path
);
1218 * pull in the delayed directory update
1219 * and the delayed inode item
1220 * otherwise we corrupt the FS during
1223 ret
= btrfs_run_delayed_items(trans
, root
);
1224 if (ret
) { /* Transaction aborted */
1225 btrfs_abort_transaction(trans
, root
, ret
);
1229 record_root_in_trans(trans
, root
);
1230 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
1231 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
1232 btrfs_check_and_init_root_item(new_root_item
);
1234 root_flags
= btrfs_root_flags(new_root_item
);
1235 if (pending
->readonly
)
1236 root_flags
|= BTRFS_ROOT_SUBVOL_RDONLY
;
1238 root_flags
&= ~BTRFS_ROOT_SUBVOL_RDONLY
;
1239 btrfs_set_root_flags(new_root_item
, root_flags
);
1241 btrfs_set_root_generation_v2(new_root_item
,
1243 uuid_le_gen(&new_uuid
);
1244 memcpy(new_root_item
->uuid
, new_uuid
.b
, BTRFS_UUID_SIZE
);
1245 memcpy(new_root_item
->parent_uuid
, root
->root_item
.uuid
,
1247 if (!(root_flags
& BTRFS_ROOT_SUBVOL_RDONLY
)) {
1248 memset(new_root_item
->received_uuid
, 0,
1249 sizeof(new_root_item
->received_uuid
));
1250 memset(&new_root_item
->stime
, 0, sizeof(new_root_item
->stime
));
1251 memset(&new_root_item
->rtime
, 0, sizeof(new_root_item
->rtime
));
1252 btrfs_set_root_stransid(new_root_item
, 0);
1253 btrfs_set_root_rtransid(new_root_item
, 0);
1255 btrfs_set_stack_timespec_sec(&new_root_item
->otime
, cur_time
.tv_sec
);
1256 btrfs_set_stack_timespec_nsec(&new_root_item
->otime
, cur_time
.tv_nsec
);
1257 btrfs_set_root_otransid(new_root_item
, trans
->transid
);
1259 old
= btrfs_lock_root_node(root
);
1260 ret
= btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
);
1262 btrfs_tree_unlock(old
);
1263 free_extent_buffer(old
);
1264 btrfs_abort_transaction(trans
, root
, ret
);
1268 btrfs_set_lock_blocking(old
);
1270 ret
= btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
1271 /* clean up in any case */
1272 btrfs_tree_unlock(old
);
1273 free_extent_buffer(old
);
1275 btrfs_abort_transaction(trans
, root
, ret
);
1279 /* see comments in should_cow_block() */
1280 root
->force_cow
= 1;
1283 btrfs_set_root_node(new_root_item
, tmp
);
1284 /* record when the snapshot was created in key.offset */
1285 key
.offset
= trans
->transid
;
1286 ret
= btrfs_insert_root(trans
, tree_root
, &key
, new_root_item
);
1287 btrfs_tree_unlock(tmp
);
1288 free_extent_buffer(tmp
);
1290 btrfs_abort_transaction(trans
, root
, ret
);
1295 * insert root back/forward references
1297 ret
= btrfs_add_root_ref(trans
, tree_root
, objectid
,
1298 parent_root
->root_key
.objectid
,
1299 btrfs_ino(parent_inode
), index
,
1300 dentry
->d_name
.name
, dentry
->d_name
.len
);
1302 btrfs_abort_transaction(trans
, root
, ret
);
1306 key
.offset
= (u64
)-1;
1307 pending
->snap
= btrfs_read_fs_root_no_name(root
->fs_info
, &key
);
1308 if (IS_ERR(pending
->snap
)) {
1309 ret
= PTR_ERR(pending
->snap
);
1310 btrfs_abort_transaction(trans
, root
, ret
);
1314 ret
= btrfs_reloc_post_snapshot(trans
, pending
);
1316 btrfs_abort_transaction(trans
, root
, ret
);
1320 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1322 btrfs_abort_transaction(trans
, root
, ret
);
1326 ret
= btrfs_insert_dir_item(trans
, parent_root
,
1327 dentry
->d_name
.name
, dentry
->d_name
.len
,
1329 BTRFS_FT_DIR
, index
);
1330 /* We have check then name at the beginning, so it is impossible. */
1331 BUG_ON(ret
== -EEXIST
|| ret
== -EOVERFLOW
);
1333 btrfs_abort_transaction(trans
, root
, ret
);
1337 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
1338 dentry
->d_name
.len
* 2);
1339 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
1340 ret
= btrfs_update_inode_fallback(trans
, parent_root
, parent_inode
);
1342 btrfs_abort_transaction(trans
, root
, ret
);
1345 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
, new_uuid
.b
,
1346 BTRFS_UUID_KEY_SUBVOL
, objectid
);
1348 btrfs_abort_transaction(trans
, root
, ret
);
1351 if (!btrfs_is_empty_uuid(new_root_item
->received_uuid
)) {
1352 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
1353 new_root_item
->received_uuid
,
1354 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
1356 if (ret
&& ret
!= -EEXIST
) {
1357 btrfs_abort_transaction(trans
, root
, ret
);
1362 pending
->error
= ret
;
1364 trans
->block_rsv
= rsv
;
1365 trans
->bytes_reserved
= 0;
1367 kfree(new_root_item
);
1368 root_item_alloc_fail
:
1369 btrfs_free_path(path
);
1374 * create all the snapshots we've scheduled for creation
1376 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
,
1377 struct btrfs_fs_info
*fs_info
)
1379 struct btrfs_pending_snapshot
*pending
, *next
;
1380 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
1383 list_for_each_entry_safe(pending
, next
, head
, list
) {
1384 list_del(&pending
->list
);
1385 ret
= create_pending_snapshot(trans
, fs_info
, pending
);
1392 static void update_super_roots(struct btrfs_root
*root
)
1394 struct btrfs_root_item
*root_item
;
1395 struct btrfs_super_block
*super
;
1397 super
= root
->fs_info
->super_copy
;
1399 root_item
= &root
->fs_info
->chunk_root
->root_item
;
1400 super
->chunk_root
= root_item
->bytenr
;
1401 super
->chunk_root_generation
= root_item
->generation
;
1402 super
->chunk_root_level
= root_item
->level
;
1404 root_item
= &root
->fs_info
->tree_root
->root_item
;
1405 super
->root
= root_item
->bytenr
;
1406 super
->generation
= root_item
->generation
;
1407 super
->root_level
= root_item
->level
;
1408 if (btrfs_test_opt(root
, SPACE_CACHE
))
1409 super
->cache_generation
= root_item
->generation
;
1410 if (root
->fs_info
->update_uuid_tree_gen
)
1411 super
->uuid_tree_generation
= root_item
->generation
;
1414 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
1416 struct btrfs_transaction
*trans
;
1419 spin_lock(&info
->trans_lock
);
1420 trans
= info
->running_transaction
;
1422 ret
= (trans
->state
>= TRANS_STATE_COMMIT_START
);
1423 spin_unlock(&info
->trans_lock
);
1427 int btrfs_transaction_blocked(struct btrfs_fs_info
*info
)
1429 struct btrfs_transaction
*trans
;
1432 spin_lock(&info
->trans_lock
);
1433 trans
= info
->running_transaction
;
1435 ret
= is_transaction_blocked(trans
);
1436 spin_unlock(&info
->trans_lock
);
1441 * wait for the current transaction commit to start and block subsequent
1444 static void wait_current_trans_commit_start(struct btrfs_root
*root
,
1445 struct btrfs_transaction
*trans
)
1447 wait_event(root
->fs_info
->transaction_blocked_wait
,
1448 trans
->state
>= TRANS_STATE_COMMIT_START
||
1453 * wait for the current transaction to start and then become unblocked.
1456 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root
*root
,
1457 struct btrfs_transaction
*trans
)
1459 wait_event(root
->fs_info
->transaction_wait
,
1460 trans
->state
>= TRANS_STATE_UNBLOCKED
||
1465 * commit transactions asynchronously. once btrfs_commit_transaction_async
1466 * returns, any subsequent transaction will not be allowed to join.
1468 struct btrfs_async_commit
{
1469 struct btrfs_trans_handle
*newtrans
;
1470 struct btrfs_root
*root
;
1471 struct work_struct work
;
1474 static void do_async_commit(struct work_struct
*work
)
1476 struct btrfs_async_commit
*ac
=
1477 container_of(work
, struct btrfs_async_commit
, work
);
1480 * We've got freeze protection passed with the transaction.
1481 * Tell lockdep about it.
1483 if (ac
->newtrans
->type
& __TRANS_FREEZABLE
)
1485 &ac
->root
->fs_info
->sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
1488 current
->journal_info
= ac
->newtrans
;
1490 btrfs_commit_transaction(ac
->newtrans
, ac
->root
);
1494 int btrfs_commit_transaction_async(struct btrfs_trans_handle
*trans
,
1495 struct btrfs_root
*root
,
1496 int wait_for_unblock
)
1498 struct btrfs_async_commit
*ac
;
1499 struct btrfs_transaction
*cur_trans
;
1501 ac
= kmalloc(sizeof(*ac
), GFP_NOFS
);
1505 INIT_WORK(&ac
->work
, do_async_commit
);
1507 ac
->newtrans
= btrfs_join_transaction(root
);
1508 if (IS_ERR(ac
->newtrans
)) {
1509 int err
= PTR_ERR(ac
->newtrans
);
1514 /* take transaction reference */
1515 cur_trans
= trans
->transaction
;
1516 atomic_inc(&cur_trans
->use_count
);
1518 btrfs_end_transaction(trans
, root
);
1521 * Tell lockdep we've released the freeze rwsem, since the
1522 * async commit thread will be the one to unlock it.
1524 if (ac
->newtrans
->type
& __TRANS_FREEZABLE
)
1526 &root
->fs_info
->sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
1529 schedule_work(&ac
->work
);
1531 /* wait for transaction to start and unblock */
1532 if (wait_for_unblock
)
1533 wait_current_trans_commit_start_and_unblock(root
, cur_trans
);
1535 wait_current_trans_commit_start(root
, cur_trans
);
1537 if (current
->journal_info
== trans
)
1538 current
->journal_info
= NULL
;
1540 btrfs_put_transaction(cur_trans
);
1545 static void cleanup_transaction(struct btrfs_trans_handle
*trans
,
1546 struct btrfs_root
*root
, int err
)
1548 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1551 WARN_ON(trans
->use_count
> 1);
1553 btrfs_abort_transaction(trans
, root
, err
);
1555 spin_lock(&root
->fs_info
->trans_lock
);
1558 * If the transaction is removed from the list, it means this
1559 * transaction has been committed successfully, so it is impossible
1560 * to call the cleanup function.
1562 BUG_ON(list_empty(&cur_trans
->list
));
1564 list_del_init(&cur_trans
->list
);
1565 if (cur_trans
== root
->fs_info
->running_transaction
) {
1566 cur_trans
->state
= TRANS_STATE_COMMIT_DOING
;
1567 spin_unlock(&root
->fs_info
->trans_lock
);
1568 wait_event(cur_trans
->writer_wait
,
1569 atomic_read(&cur_trans
->num_writers
) == 1);
1571 spin_lock(&root
->fs_info
->trans_lock
);
1573 spin_unlock(&root
->fs_info
->trans_lock
);
1575 btrfs_cleanup_one_transaction(trans
->transaction
, root
);
1577 spin_lock(&root
->fs_info
->trans_lock
);
1578 if (cur_trans
== root
->fs_info
->running_transaction
)
1579 root
->fs_info
->running_transaction
= NULL
;
1580 spin_unlock(&root
->fs_info
->trans_lock
);
1582 if (trans
->type
& __TRANS_FREEZABLE
)
1583 sb_end_intwrite(root
->fs_info
->sb
);
1584 btrfs_put_transaction(cur_trans
);
1585 btrfs_put_transaction(cur_trans
);
1587 trace_btrfs_transaction_commit(root
);
1589 btrfs_scrub_continue(root
);
1591 if (current
->journal_info
== trans
)
1592 current
->journal_info
= NULL
;
1594 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1597 static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle
*trans
,
1598 struct btrfs_root
*root
)
1602 ret
= btrfs_run_delayed_items(trans
, root
);
1604 * running the delayed items may have added new refs. account
1605 * them now so that they hinder processing of more delayed refs
1606 * as little as possible.
1609 btrfs_delayed_refs_qgroup_accounting(trans
, root
->fs_info
);
1613 ret
= btrfs_delayed_refs_qgroup_accounting(trans
, root
->fs_info
);
1618 * rename don't use btrfs_join_transaction, so, once we
1619 * set the transaction to blocked above, we aren't going
1620 * to get any new ordered operations. We can safely run
1621 * it here and no for sure that nothing new will be added
1624 ret
= btrfs_run_ordered_operations(trans
, root
, 1);
1629 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info
*fs_info
)
1631 if (btrfs_test_opt(fs_info
->tree_root
, FLUSHONCOMMIT
))
1632 return btrfs_start_delalloc_roots(fs_info
, 1);
1636 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info
*fs_info
)
1638 if (btrfs_test_opt(fs_info
->tree_root
, FLUSHONCOMMIT
))
1639 btrfs_wait_ordered_roots(fs_info
, -1);
1642 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
,
1643 struct btrfs_root
*root
)
1645 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1646 struct btrfs_transaction
*prev_trans
= NULL
;
1649 ret
= btrfs_run_ordered_operations(trans
, root
, 0);
1651 btrfs_abort_transaction(trans
, root
, ret
);
1652 btrfs_end_transaction(trans
, root
);
1656 /* Stop the commit early if ->aborted is set */
1657 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1658 ret
= cur_trans
->aborted
;
1659 btrfs_end_transaction(trans
, root
);
1663 /* make a pass through all the delayed refs we have so far
1664 * any runnings procs may add more while we are here
1666 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1668 btrfs_end_transaction(trans
, root
);
1672 btrfs_trans_release_metadata(trans
, root
);
1673 trans
->block_rsv
= NULL
;
1674 if (trans
->qgroup_reserved
) {
1675 btrfs_qgroup_free(root
, trans
->qgroup_reserved
);
1676 trans
->qgroup_reserved
= 0;
1679 cur_trans
= trans
->transaction
;
1682 * set the flushing flag so procs in this transaction have to
1683 * start sending their work down.
1685 cur_trans
->delayed_refs
.flushing
= 1;
1688 if (!list_empty(&trans
->new_bgs
))
1689 btrfs_create_pending_block_groups(trans
, root
);
1691 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1693 btrfs_end_transaction(trans
, root
);
1697 spin_lock(&root
->fs_info
->trans_lock
);
1698 if (cur_trans
->state
>= TRANS_STATE_COMMIT_START
) {
1699 spin_unlock(&root
->fs_info
->trans_lock
);
1700 atomic_inc(&cur_trans
->use_count
);
1701 ret
= btrfs_end_transaction(trans
, root
);
1703 wait_for_commit(root
, cur_trans
);
1705 btrfs_put_transaction(cur_trans
);
1710 cur_trans
->state
= TRANS_STATE_COMMIT_START
;
1711 wake_up(&root
->fs_info
->transaction_blocked_wait
);
1713 if (cur_trans
->list
.prev
!= &root
->fs_info
->trans_list
) {
1714 prev_trans
= list_entry(cur_trans
->list
.prev
,
1715 struct btrfs_transaction
, list
);
1716 if (prev_trans
->state
!= TRANS_STATE_COMPLETED
) {
1717 atomic_inc(&prev_trans
->use_count
);
1718 spin_unlock(&root
->fs_info
->trans_lock
);
1720 wait_for_commit(root
, prev_trans
);
1722 btrfs_put_transaction(prev_trans
);
1724 spin_unlock(&root
->fs_info
->trans_lock
);
1727 spin_unlock(&root
->fs_info
->trans_lock
);
1730 extwriter_counter_dec(cur_trans
, trans
->type
);
1732 ret
= btrfs_start_delalloc_flush(root
->fs_info
);
1734 goto cleanup_transaction
;
1736 ret
= btrfs_flush_all_pending_stuffs(trans
, root
);
1738 goto cleanup_transaction
;
1740 wait_event(cur_trans
->writer_wait
,
1741 extwriter_counter_read(cur_trans
) == 0);
1743 /* some pending stuffs might be added after the previous flush. */
1744 ret
= btrfs_flush_all_pending_stuffs(trans
, root
);
1746 goto cleanup_transaction
;
1748 btrfs_wait_delalloc_flush(root
->fs_info
);
1750 * Ok now we need to make sure to block out any other joins while we
1751 * commit the transaction. We could have started a join before setting
1752 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
1754 spin_lock(&root
->fs_info
->trans_lock
);
1755 cur_trans
->state
= TRANS_STATE_COMMIT_DOING
;
1756 spin_unlock(&root
->fs_info
->trans_lock
);
1757 wait_event(cur_trans
->writer_wait
,
1758 atomic_read(&cur_trans
->num_writers
) == 1);
1760 /* ->aborted might be set after the previous check, so check it */
1761 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1762 ret
= cur_trans
->aborted
;
1763 goto cleanup_transaction
;
1766 * the reloc mutex makes sure that we stop
1767 * the balancing code from coming in and moving
1768 * extents around in the middle of the commit
1770 mutex_lock(&root
->fs_info
->reloc_mutex
);
1773 * We needn't worry about the delayed items because we will
1774 * deal with them in create_pending_snapshot(), which is the
1775 * core function of the snapshot creation.
1777 ret
= create_pending_snapshots(trans
, root
->fs_info
);
1779 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1780 goto cleanup_transaction
;
1784 * We insert the dir indexes of the snapshots and update the inode
1785 * of the snapshots' parents after the snapshot creation, so there
1786 * are some delayed items which are not dealt with. Now deal with
1789 * We needn't worry that this operation will corrupt the snapshots,
1790 * because all the tree which are snapshoted will be forced to COW
1791 * the nodes and leaves.
1793 ret
= btrfs_run_delayed_items(trans
, root
);
1795 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1796 goto cleanup_transaction
;
1799 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1801 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1802 goto cleanup_transaction
;
1806 * make sure none of the code above managed to slip in a
1809 btrfs_assert_delayed_root_empty(root
);
1811 WARN_ON(cur_trans
!= trans
->transaction
);
1813 btrfs_scrub_pause(root
);
1814 /* btrfs_commit_tree_roots is responsible for getting the
1815 * various roots consistent with each other. Every pointer
1816 * in the tree of tree roots has to point to the most up to date
1817 * root for every subvolume and other tree. So, we have to keep
1818 * the tree logging code from jumping in and changing any
1821 * At this point in the commit, there can't be any tree-log
1822 * writers, but a little lower down we drop the trans mutex
1823 * and let new people in. By holding the tree_log_mutex
1824 * from now until after the super is written, we avoid races
1825 * with the tree-log code.
1827 mutex_lock(&root
->fs_info
->tree_log_mutex
);
1829 ret
= commit_fs_roots(trans
, root
);
1831 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1832 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1833 goto cleanup_transaction
;
1836 /* commit_fs_roots gets rid of all the tree log roots, it is now
1837 * safe to free the root of tree log roots
1839 btrfs_free_log_root_tree(trans
, root
->fs_info
);
1841 ret
= commit_cowonly_roots(trans
, root
);
1843 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1844 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1845 goto cleanup_transaction
;
1849 * The tasks which save the space cache and inode cache may also
1850 * update ->aborted, check it.
1852 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1853 ret
= cur_trans
->aborted
;
1854 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1855 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1856 goto cleanup_transaction
;
1859 btrfs_prepare_extent_commit(trans
, root
);
1861 cur_trans
= root
->fs_info
->running_transaction
;
1863 btrfs_set_root_node(&root
->fs_info
->tree_root
->root_item
,
1864 root
->fs_info
->tree_root
->node
);
1865 switch_commit_root(root
->fs_info
->tree_root
);
1867 btrfs_set_root_node(&root
->fs_info
->chunk_root
->root_item
,
1868 root
->fs_info
->chunk_root
->node
);
1869 switch_commit_root(root
->fs_info
->chunk_root
);
1871 assert_qgroups_uptodate(trans
);
1872 update_super_roots(root
);
1874 btrfs_set_super_log_root(root
->fs_info
->super_copy
, 0);
1875 btrfs_set_super_log_root_level(root
->fs_info
->super_copy
, 0);
1876 memcpy(root
->fs_info
->super_for_commit
, root
->fs_info
->super_copy
,
1877 sizeof(*root
->fs_info
->super_copy
));
1879 spin_lock(&root
->fs_info
->trans_lock
);
1880 cur_trans
->state
= TRANS_STATE_UNBLOCKED
;
1881 root
->fs_info
->running_transaction
= NULL
;
1882 spin_unlock(&root
->fs_info
->trans_lock
);
1883 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1885 wake_up(&root
->fs_info
->transaction_wait
);
1887 ret
= btrfs_write_and_wait_transaction(trans
, root
);
1889 btrfs_error(root
->fs_info
, ret
,
1890 "Error while writing out transaction");
1891 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1892 goto cleanup_transaction
;
1895 ret
= write_ctree_super(trans
, root
, 0);
1897 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1898 goto cleanup_transaction
;
1902 * the super is written, we can safely allow the tree-loggers
1903 * to go about their business
1905 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1907 btrfs_finish_extent_commit(trans
, root
);
1909 root
->fs_info
->last_trans_committed
= cur_trans
->transid
;
1911 * We needn't acquire the lock here because there is no other task
1912 * which can change it.
1914 cur_trans
->state
= TRANS_STATE_COMPLETED
;
1915 wake_up(&cur_trans
->commit_wait
);
1917 spin_lock(&root
->fs_info
->trans_lock
);
1918 list_del_init(&cur_trans
->list
);
1919 spin_unlock(&root
->fs_info
->trans_lock
);
1921 btrfs_put_transaction(cur_trans
);
1922 btrfs_put_transaction(cur_trans
);
1924 if (trans
->type
& __TRANS_FREEZABLE
)
1925 sb_end_intwrite(root
->fs_info
->sb
);
1927 trace_btrfs_transaction_commit(root
);
1929 btrfs_scrub_continue(root
);
1931 if (current
->journal_info
== trans
)
1932 current
->journal_info
= NULL
;
1934 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1936 if (current
!= root
->fs_info
->transaction_kthread
)
1937 btrfs_run_delayed_iputs(root
);
1941 cleanup_transaction
:
1942 btrfs_trans_release_metadata(trans
, root
);
1943 trans
->block_rsv
= NULL
;
1944 if (trans
->qgroup_reserved
) {
1945 btrfs_qgroup_free(root
, trans
->qgroup_reserved
);
1946 trans
->qgroup_reserved
= 0;
1948 btrfs_warn(root
->fs_info
, "Skipping commit of aborted transaction.");
1949 if (current
->journal_info
== trans
)
1950 current
->journal_info
= NULL
;
1951 cleanup_transaction(trans
, root
, ret
);
1957 * return < 0 if error
1958 * 0 if there are no more dead_roots at the time of call
1959 * 1 there are more to be processed, call me again
1961 * The return value indicates there are certainly more snapshots to delete, but
1962 * if there comes a new one during processing, it may return 0. We don't mind,
1963 * because btrfs_commit_super will poke cleaner thread and it will process it a
1964 * few seconds later.
1966 int btrfs_clean_one_deleted_snapshot(struct btrfs_root
*root
)
1969 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1971 spin_lock(&fs_info
->trans_lock
);
1972 if (list_empty(&fs_info
->dead_roots
)) {
1973 spin_unlock(&fs_info
->trans_lock
);
1976 root
= list_first_entry(&fs_info
->dead_roots
,
1977 struct btrfs_root
, root_list
);
1978 list_del_init(&root
->root_list
);
1979 spin_unlock(&fs_info
->trans_lock
);
1981 pr_debug("btrfs: cleaner removing %llu\n", root
->objectid
);
1983 btrfs_kill_all_delayed_nodes(root
);
1985 if (btrfs_header_backref_rev(root
->node
) <
1986 BTRFS_MIXED_BACKREF_REV
)
1987 ret
= btrfs_drop_snapshot(root
, NULL
, 0, 0);
1989 ret
= btrfs_drop_snapshot(root
, NULL
, 1, 0);
1991 * If we encounter a transaction abort during snapshot cleaning, we
1992 * don't want to crash here
1994 return (ret
< 0) ? 0 : 1;