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(!RB_EMPTY_ROOT(&transaction
->delayed_refs
.root
));
66 WARN_ON(!RB_EMPTY_ROOT(&transaction
->delayed_refs
.href_root
));
67 while (!list_empty(&transaction
->pending_chunks
)) {
68 struct extent_map
*em
;
70 em
= list_first_entry(&transaction
->pending_chunks
,
71 struct extent_map
, list
);
72 list_del_init(&em
->list
);
75 kmem_cache_free(btrfs_transaction_cachep
, transaction
);
79 static noinline
void switch_commit_root(struct btrfs_root
*root
)
81 free_extent_buffer(root
->commit_root
);
82 root
->commit_root
= btrfs_root_node(root
);
85 static inline void extwriter_counter_inc(struct btrfs_transaction
*trans
,
88 if (type
& TRANS_EXTWRITERS
)
89 atomic_inc(&trans
->num_extwriters
);
92 static inline void extwriter_counter_dec(struct btrfs_transaction
*trans
,
95 if (type
& TRANS_EXTWRITERS
)
96 atomic_dec(&trans
->num_extwriters
);
99 static inline void extwriter_counter_init(struct btrfs_transaction
*trans
,
102 atomic_set(&trans
->num_extwriters
, ((type
& TRANS_EXTWRITERS
) ? 1 : 0));
105 static inline int extwriter_counter_read(struct btrfs_transaction
*trans
)
107 return atomic_read(&trans
->num_extwriters
);
111 * either allocate a new transaction or hop into the existing one
113 static noinline
int join_transaction(struct btrfs_root
*root
, unsigned int type
)
115 struct btrfs_transaction
*cur_trans
;
116 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
118 spin_lock(&fs_info
->trans_lock
);
120 /* The file system has been taken offline. No new transactions. */
121 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
122 spin_unlock(&fs_info
->trans_lock
);
126 cur_trans
= fs_info
->running_transaction
;
128 if (cur_trans
->aborted
) {
129 spin_unlock(&fs_info
->trans_lock
);
130 return cur_trans
->aborted
;
132 if (btrfs_blocked_trans_types
[cur_trans
->state
] & type
) {
133 spin_unlock(&fs_info
->trans_lock
);
136 atomic_inc(&cur_trans
->use_count
);
137 atomic_inc(&cur_trans
->num_writers
);
138 extwriter_counter_inc(cur_trans
, type
);
139 spin_unlock(&fs_info
->trans_lock
);
142 spin_unlock(&fs_info
->trans_lock
);
145 * If we are ATTACH, we just want to catch the current transaction,
146 * and commit it. If there is no transaction, just return ENOENT.
148 if (type
== TRANS_ATTACH
)
152 * JOIN_NOLOCK only happens during the transaction commit, so
153 * it is impossible that ->running_transaction is NULL
155 BUG_ON(type
== TRANS_JOIN_NOLOCK
);
157 cur_trans
= kmem_cache_alloc(btrfs_transaction_cachep
, GFP_NOFS
);
161 spin_lock(&fs_info
->trans_lock
);
162 if (fs_info
->running_transaction
) {
164 * someone started a transaction after we unlocked. Make sure
165 * to redo the checks above
167 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
169 } else if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
170 spin_unlock(&fs_info
->trans_lock
);
171 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
175 atomic_set(&cur_trans
->num_writers
, 1);
176 extwriter_counter_init(cur_trans
, type
);
177 init_waitqueue_head(&cur_trans
->writer_wait
);
178 init_waitqueue_head(&cur_trans
->commit_wait
);
179 cur_trans
->state
= TRANS_STATE_RUNNING
;
181 * One for this trans handle, one so it will live on until we
182 * commit the transaction.
184 atomic_set(&cur_trans
->use_count
, 2);
185 cur_trans
->start_time
= get_seconds();
187 cur_trans
->delayed_refs
.root
= RB_ROOT
;
188 cur_trans
->delayed_refs
.href_root
= RB_ROOT
;
189 cur_trans
->delayed_refs
.num_entries
= 0;
190 cur_trans
->delayed_refs
.num_heads_ready
= 0;
191 cur_trans
->delayed_refs
.num_heads
= 0;
192 cur_trans
->delayed_refs
.flushing
= 0;
193 cur_trans
->delayed_refs
.run_delayed_start
= 0;
196 * although the tree mod log is per file system and not per transaction,
197 * the log must never go across transaction boundaries.
200 if (!list_empty(&fs_info
->tree_mod_seq_list
))
201 WARN(1, KERN_ERR
"BTRFS: tree_mod_seq_list not empty when "
202 "creating a fresh transaction\n");
203 if (!RB_EMPTY_ROOT(&fs_info
->tree_mod_log
))
204 WARN(1, KERN_ERR
"BTRFS: tree_mod_log rb tree not empty when "
205 "creating a fresh transaction\n");
206 atomic64_set(&fs_info
->tree_mod_seq
, 0);
208 spin_lock_init(&cur_trans
->delayed_refs
.lock
);
209 atomic_set(&cur_trans
->delayed_refs
.procs_running_refs
, 0);
210 atomic_set(&cur_trans
->delayed_refs
.ref_seq
, 0);
211 init_waitqueue_head(&cur_trans
->delayed_refs
.wait
);
213 INIT_LIST_HEAD(&cur_trans
->pending_snapshots
);
214 INIT_LIST_HEAD(&cur_trans
->ordered_operations
);
215 INIT_LIST_HEAD(&cur_trans
->pending_chunks
);
216 list_add_tail(&cur_trans
->list
, &fs_info
->trans_list
);
217 extent_io_tree_init(&cur_trans
->dirty_pages
,
218 fs_info
->btree_inode
->i_mapping
);
219 fs_info
->generation
++;
220 cur_trans
->transid
= fs_info
->generation
;
221 fs_info
->running_transaction
= cur_trans
;
222 cur_trans
->aborted
= 0;
223 spin_unlock(&fs_info
->trans_lock
);
229 * this does all the record keeping required to make sure that a reference
230 * counted root is properly recorded in a given transaction. This is required
231 * to make sure the old root from before we joined the transaction is deleted
232 * when the transaction commits
234 static int record_root_in_trans(struct btrfs_trans_handle
*trans
,
235 struct btrfs_root
*root
)
237 if (root
->ref_cows
&& root
->last_trans
< trans
->transid
) {
238 WARN_ON(root
== root
->fs_info
->extent_root
);
239 WARN_ON(root
->commit_root
!= root
->node
);
242 * see below for in_trans_setup usage rules
243 * we have the reloc mutex held now, so there
244 * is only one writer in this function
246 root
->in_trans_setup
= 1;
248 /* make sure readers find in_trans_setup before
249 * they find our root->last_trans update
253 spin_lock(&root
->fs_info
->fs_roots_radix_lock
);
254 if (root
->last_trans
== trans
->transid
) {
255 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
258 radix_tree_tag_set(&root
->fs_info
->fs_roots_radix
,
259 (unsigned long)root
->root_key
.objectid
,
260 BTRFS_ROOT_TRANS_TAG
);
261 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
262 root
->last_trans
= trans
->transid
;
264 /* this is pretty tricky. We don't want to
265 * take the relocation lock in btrfs_record_root_in_trans
266 * unless we're really doing the first setup for this root in
269 * Normally we'd use root->last_trans as a flag to decide
270 * if we want to take the expensive mutex.
272 * But, we have to set root->last_trans before we
273 * init the relocation root, otherwise, we trip over warnings
274 * in ctree.c. The solution used here is to flag ourselves
275 * with root->in_trans_setup. When this is 1, we're still
276 * fixing up the reloc trees and everyone must wait.
278 * When this is zero, they can trust root->last_trans and fly
279 * through btrfs_record_root_in_trans without having to take the
280 * lock. smp_wmb() makes sure that all the writes above are
281 * done before we pop in the zero below
283 btrfs_init_reloc_root(trans
, root
);
285 root
->in_trans_setup
= 0;
291 int btrfs_record_root_in_trans(struct btrfs_trans_handle
*trans
,
292 struct btrfs_root
*root
)
298 * see record_root_in_trans for comments about in_trans_setup usage
302 if (root
->last_trans
== trans
->transid
&&
303 !root
->in_trans_setup
)
306 mutex_lock(&root
->fs_info
->reloc_mutex
);
307 record_root_in_trans(trans
, root
);
308 mutex_unlock(&root
->fs_info
->reloc_mutex
);
313 static inline int is_transaction_blocked(struct btrfs_transaction
*trans
)
315 return (trans
->state
>= TRANS_STATE_BLOCKED
&&
316 trans
->state
< TRANS_STATE_UNBLOCKED
&&
320 /* wait for commit against the current transaction to become unblocked
321 * when this is done, it is safe to start a new transaction, but the current
322 * transaction might not be fully on disk.
324 static void wait_current_trans(struct btrfs_root
*root
)
326 struct btrfs_transaction
*cur_trans
;
328 spin_lock(&root
->fs_info
->trans_lock
);
329 cur_trans
= root
->fs_info
->running_transaction
;
330 if (cur_trans
&& is_transaction_blocked(cur_trans
)) {
331 atomic_inc(&cur_trans
->use_count
);
332 spin_unlock(&root
->fs_info
->trans_lock
);
334 wait_event(root
->fs_info
->transaction_wait
,
335 cur_trans
->state
>= TRANS_STATE_UNBLOCKED
||
337 btrfs_put_transaction(cur_trans
);
339 spin_unlock(&root
->fs_info
->trans_lock
);
343 static int may_wait_transaction(struct btrfs_root
*root
, int type
)
345 if (root
->fs_info
->log_root_recovering
)
348 if (type
== TRANS_USERSPACE
)
351 if (type
== TRANS_START
&&
352 !atomic_read(&root
->fs_info
->open_ioctl_trans
))
358 static inline bool need_reserve_reloc_root(struct btrfs_root
*root
)
360 if (!root
->fs_info
->reloc_ctl
||
362 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
369 static struct btrfs_trans_handle
*
370 start_transaction(struct btrfs_root
*root
, u64 num_items
, unsigned int type
,
371 enum btrfs_reserve_flush_enum flush
)
373 struct btrfs_trans_handle
*h
;
374 struct btrfs_transaction
*cur_trans
;
376 u64 qgroup_reserved
= 0;
377 bool reloc_reserved
= false;
380 if (test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
381 return ERR_PTR(-EROFS
);
383 if (current
->journal_info
) {
384 WARN_ON(type
& TRANS_EXTWRITERS
);
385 h
= current
->journal_info
;
387 WARN_ON(h
->use_count
> 2);
388 h
->orig_rsv
= h
->block_rsv
;
394 * Do the reservation before we join the transaction so we can do all
395 * the appropriate flushing if need be.
397 if (num_items
> 0 && root
!= root
->fs_info
->chunk_root
) {
398 if (root
->fs_info
->quota_enabled
&&
399 is_fstree(root
->root_key
.objectid
)) {
400 qgroup_reserved
= num_items
* root
->leafsize
;
401 ret
= btrfs_qgroup_reserve(root
, qgroup_reserved
);
406 num_bytes
= btrfs_calc_trans_metadata_size(root
, num_items
);
408 * Do the reservation for the relocation root creation
410 if (unlikely(need_reserve_reloc_root(root
))) {
411 num_bytes
+= root
->nodesize
;
412 reloc_reserved
= true;
415 ret
= btrfs_block_rsv_add(root
,
416 &root
->fs_info
->trans_block_rsv
,
422 h
= kmem_cache_alloc(btrfs_trans_handle_cachep
, GFP_NOFS
);
429 * If we are JOIN_NOLOCK we're already committing a transaction and
430 * waiting on this guy, so we don't need to do the sb_start_intwrite
431 * because we're already holding a ref. We need this because we could
432 * have raced in and did an fsync() on a file which can kick a commit
433 * and then we deadlock with somebody doing a freeze.
435 * If we are ATTACH, it means we just want to catch the current
436 * transaction and commit it, so we needn't do sb_start_intwrite().
438 if (type
& __TRANS_FREEZABLE
)
439 sb_start_intwrite(root
->fs_info
->sb
);
441 if (may_wait_transaction(root
, type
))
442 wait_current_trans(root
);
445 ret
= join_transaction(root
, type
);
447 wait_current_trans(root
);
448 if (unlikely(type
== TRANS_ATTACH
))
451 } while (ret
== -EBUSY
);
454 /* We must get the transaction if we are JOIN_NOLOCK. */
455 BUG_ON(type
== TRANS_JOIN_NOLOCK
);
459 cur_trans
= root
->fs_info
->running_transaction
;
461 h
->transid
= cur_trans
->transid
;
462 h
->transaction
= cur_trans
;
464 h
->bytes_reserved
= 0;
466 h
->delayed_ref_updates
= 0;
472 h
->qgroup_reserved
= 0;
473 h
->delayed_ref_elem
.seq
= 0;
475 h
->allocating_chunk
= false;
476 h
->reloc_reserved
= false;
477 INIT_LIST_HEAD(&h
->qgroup_ref_list
);
478 INIT_LIST_HEAD(&h
->new_bgs
);
481 if (cur_trans
->state
>= TRANS_STATE_BLOCKED
&&
482 may_wait_transaction(root
, type
)) {
483 btrfs_commit_transaction(h
, root
);
488 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
489 h
->transid
, num_bytes
, 1);
490 h
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
491 h
->bytes_reserved
= num_bytes
;
492 h
->reloc_reserved
= reloc_reserved
;
494 h
->qgroup_reserved
= qgroup_reserved
;
497 btrfs_record_root_in_trans(h
, root
);
499 if (!current
->journal_info
&& type
!= TRANS_USERSPACE
)
500 current
->journal_info
= h
;
504 if (type
& __TRANS_FREEZABLE
)
505 sb_end_intwrite(root
->fs_info
->sb
);
506 kmem_cache_free(btrfs_trans_handle_cachep
, h
);
509 btrfs_block_rsv_release(root
, &root
->fs_info
->trans_block_rsv
,
513 btrfs_qgroup_free(root
, qgroup_reserved
);
517 struct btrfs_trans_handle
*btrfs_start_transaction(struct btrfs_root
*root
,
520 return start_transaction(root
, num_items
, TRANS_START
,
521 BTRFS_RESERVE_FLUSH_ALL
);
524 struct btrfs_trans_handle
*btrfs_start_transaction_lflush(
525 struct btrfs_root
*root
, int num_items
)
527 return start_transaction(root
, num_items
, TRANS_START
,
528 BTRFS_RESERVE_FLUSH_LIMIT
);
531 struct btrfs_trans_handle
*btrfs_join_transaction(struct btrfs_root
*root
)
533 return start_transaction(root
, 0, TRANS_JOIN
, 0);
536 struct btrfs_trans_handle
*btrfs_join_transaction_nolock(struct btrfs_root
*root
)
538 return start_transaction(root
, 0, TRANS_JOIN_NOLOCK
, 0);
541 struct btrfs_trans_handle
*btrfs_start_ioctl_transaction(struct btrfs_root
*root
)
543 return start_transaction(root
, 0, TRANS_USERSPACE
, 0);
547 * btrfs_attach_transaction() - catch the running transaction
549 * It is used when we want to commit the current the transaction, but
550 * don't want to start a new one.
552 * Note: If this function return -ENOENT, it just means there is no
553 * running transaction. But it is possible that the inactive transaction
554 * is still in the memory, not fully on disk. If you hope there is no
555 * inactive transaction in the fs when -ENOENT is returned, you should
557 * btrfs_attach_transaction_barrier()
559 struct btrfs_trans_handle
*btrfs_attach_transaction(struct btrfs_root
*root
)
561 return start_transaction(root
, 0, TRANS_ATTACH
, 0);
565 * btrfs_attach_transaction_barrier() - catch the running transaction
567 * It is similar to the above function, the differentia is this one
568 * will wait for all the inactive transactions until they fully
571 struct btrfs_trans_handle
*
572 btrfs_attach_transaction_barrier(struct btrfs_root
*root
)
574 struct btrfs_trans_handle
*trans
;
576 trans
= start_transaction(root
, 0, TRANS_ATTACH
, 0);
577 if (IS_ERR(trans
) && PTR_ERR(trans
) == -ENOENT
)
578 btrfs_wait_for_commit(root
, 0);
583 /* wait for a transaction commit to be fully complete */
584 static noinline
void wait_for_commit(struct btrfs_root
*root
,
585 struct btrfs_transaction
*commit
)
587 wait_event(commit
->commit_wait
, commit
->state
== TRANS_STATE_COMPLETED
);
590 int btrfs_wait_for_commit(struct btrfs_root
*root
, u64 transid
)
592 struct btrfs_transaction
*cur_trans
= NULL
, *t
;
596 if (transid
<= root
->fs_info
->last_trans_committed
)
600 /* find specified transaction */
601 spin_lock(&root
->fs_info
->trans_lock
);
602 list_for_each_entry(t
, &root
->fs_info
->trans_list
, list
) {
603 if (t
->transid
== transid
) {
605 atomic_inc(&cur_trans
->use_count
);
609 if (t
->transid
> transid
) {
614 spin_unlock(&root
->fs_info
->trans_lock
);
615 /* The specified transaction doesn't exist */
619 /* find newest transaction that is committing | committed */
620 spin_lock(&root
->fs_info
->trans_lock
);
621 list_for_each_entry_reverse(t
, &root
->fs_info
->trans_list
,
623 if (t
->state
>= TRANS_STATE_COMMIT_START
) {
624 if (t
->state
== TRANS_STATE_COMPLETED
)
627 atomic_inc(&cur_trans
->use_count
);
631 spin_unlock(&root
->fs_info
->trans_lock
);
633 goto out
; /* nothing committing|committed */
636 wait_for_commit(root
, cur_trans
);
637 btrfs_put_transaction(cur_trans
);
642 void btrfs_throttle(struct btrfs_root
*root
)
644 if (!atomic_read(&root
->fs_info
->open_ioctl_trans
))
645 wait_current_trans(root
);
648 static int should_end_transaction(struct btrfs_trans_handle
*trans
,
649 struct btrfs_root
*root
)
651 if (root
->fs_info
->global_block_rsv
.space_info
->full
&&
652 btrfs_should_throttle_delayed_refs(trans
, root
))
655 return !!btrfs_block_rsv_check(root
, &root
->fs_info
->global_block_rsv
, 5);
658 int btrfs_should_end_transaction(struct btrfs_trans_handle
*trans
,
659 struct btrfs_root
*root
)
661 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
666 if (cur_trans
->state
>= TRANS_STATE_BLOCKED
||
667 cur_trans
->delayed_refs
.flushing
)
670 updates
= trans
->delayed_ref_updates
;
671 trans
->delayed_ref_updates
= 0;
673 err
= btrfs_run_delayed_refs(trans
, root
, updates
);
674 if (err
) /* Error code will also eval true */
678 return should_end_transaction(trans
, root
);
681 static int __btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
682 struct btrfs_root
*root
, int throttle
)
684 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
685 struct btrfs_fs_info
*info
= root
->fs_info
;
686 unsigned long cur
= trans
->delayed_ref_updates
;
687 int lock
= (trans
->type
!= TRANS_JOIN_NOLOCK
);
690 if (--trans
->use_count
) {
691 trans
->block_rsv
= trans
->orig_rsv
;
696 * do the qgroup accounting as early as possible
698 err
= btrfs_delayed_refs_qgroup_accounting(trans
, info
);
700 btrfs_trans_release_metadata(trans
, root
);
701 trans
->block_rsv
= NULL
;
703 if (trans
->qgroup_reserved
) {
705 * the same root has to be passed here between start_transaction
706 * and end_transaction. Subvolume quota depends on this.
708 btrfs_qgroup_free(trans
->root
, trans
->qgroup_reserved
);
709 trans
->qgroup_reserved
= 0;
712 if (!list_empty(&trans
->new_bgs
))
713 btrfs_create_pending_block_groups(trans
, root
);
715 trans
->delayed_ref_updates
= 0;
716 if (btrfs_should_throttle_delayed_refs(trans
, root
)) {
717 cur
= max_t(unsigned long, cur
, 1);
718 trans
->delayed_ref_updates
= 0;
719 btrfs_run_delayed_refs(trans
, root
, cur
);
722 btrfs_trans_release_metadata(trans
, root
);
723 trans
->block_rsv
= NULL
;
725 if (!list_empty(&trans
->new_bgs
))
726 btrfs_create_pending_block_groups(trans
, root
);
728 if (lock
&& !atomic_read(&root
->fs_info
->open_ioctl_trans
) &&
729 should_end_transaction(trans
, root
) &&
730 ACCESS_ONCE(cur_trans
->state
) == TRANS_STATE_RUNNING
) {
731 spin_lock(&info
->trans_lock
);
732 if (cur_trans
->state
== TRANS_STATE_RUNNING
)
733 cur_trans
->state
= TRANS_STATE_BLOCKED
;
734 spin_unlock(&info
->trans_lock
);
737 if (lock
&& ACCESS_ONCE(cur_trans
->state
) == TRANS_STATE_BLOCKED
) {
740 * We may race with somebody else here so end up having
741 * to call end_transaction on ourselves again, so inc
745 return btrfs_commit_transaction(trans
, root
);
747 wake_up_process(info
->transaction_kthread
);
751 if (trans
->type
& __TRANS_FREEZABLE
)
752 sb_end_intwrite(root
->fs_info
->sb
);
754 WARN_ON(cur_trans
!= info
->running_transaction
);
755 WARN_ON(atomic_read(&cur_trans
->num_writers
) < 1);
756 atomic_dec(&cur_trans
->num_writers
);
757 extwriter_counter_dec(cur_trans
, trans
->type
);
760 if (waitqueue_active(&cur_trans
->writer_wait
))
761 wake_up(&cur_trans
->writer_wait
);
762 btrfs_put_transaction(cur_trans
);
764 if (current
->journal_info
== trans
)
765 current
->journal_info
= NULL
;
768 btrfs_run_delayed_iputs(root
);
770 if (trans
->aborted
||
771 test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
)) {
772 wake_up_process(info
->transaction_kthread
);
775 assert_qgroups_uptodate(trans
);
777 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
781 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
782 struct btrfs_root
*root
)
784 return __btrfs_end_transaction(trans
, root
, 0);
787 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
,
788 struct btrfs_root
*root
)
790 return __btrfs_end_transaction(trans
, root
, 1);
793 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle
*trans
,
794 struct btrfs_root
*root
)
796 return __btrfs_end_transaction(trans
, root
, 1);
800 * when btree blocks are allocated, they have some corresponding bits set for
801 * them in one of two extent_io trees. This is used to make sure all of
802 * those extents are sent to disk but does not wait on them
804 int btrfs_write_marked_extents(struct btrfs_root
*root
,
805 struct extent_io_tree
*dirty_pages
, int mark
)
809 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
810 struct extent_state
*cached_state
= NULL
;
814 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
815 mark
, &cached_state
)) {
816 convert_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
,
817 mark
, &cached_state
, GFP_NOFS
);
819 err
= filemap_fdatawrite_range(mapping
, start
, end
);
831 * when btree blocks are allocated, they have some corresponding bits set for
832 * them in one of two extent_io trees. This is used to make sure all of
833 * those extents are on disk for transaction or log commit. We wait
834 * on all the pages and clear them from the dirty pages state tree
836 int btrfs_wait_marked_extents(struct btrfs_root
*root
,
837 struct extent_io_tree
*dirty_pages
, int mark
)
841 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
842 struct extent_state
*cached_state
= NULL
;
846 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
847 EXTENT_NEED_WAIT
, &cached_state
)) {
848 clear_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
,
849 0, 0, &cached_state
, GFP_NOFS
);
850 err
= filemap_fdatawait_range(mapping
, start
, end
);
862 * when btree blocks are allocated, they have some corresponding bits set for
863 * them in one of two extent_io trees. This is used to make sure all of
864 * those extents are on disk for transaction or log commit
866 static int btrfs_write_and_wait_marked_extents(struct btrfs_root
*root
,
867 struct extent_io_tree
*dirty_pages
, int mark
)
871 struct blk_plug plug
;
873 blk_start_plug(&plug
);
874 ret
= btrfs_write_marked_extents(root
, dirty_pages
, mark
);
875 blk_finish_plug(&plug
);
876 ret2
= btrfs_wait_marked_extents(root
, dirty_pages
, mark
);
885 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
,
886 struct btrfs_root
*root
)
888 if (!trans
|| !trans
->transaction
) {
889 struct inode
*btree_inode
;
890 btree_inode
= root
->fs_info
->btree_inode
;
891 return filemap_write_and_wait(btree_inode
->i_mapping
);
893 return btrfs_write_and_wait_marked_extents(root
,
894 &trans
->transaction
->dirty_pages
,
899 * this is used to update the root pointer in the tree of tree roots.
901 * But, in the case of the extent allocation tree, updating the root
902 * pointer may allocate blocks which may change the root of the extent
905 * So, this loops and repeats and makes sure the cowonly root didn't
906 * change while the root pointer was being updated in the metadata.
908 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
909 struct btrfs_root
*root
)
914 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
916 old_root_used
= btrfs_root_used(&root
->root_item
);
917 btrfs_write_dirty_block_groups(trans
, root
);
920 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
921 if (old_root_bytenr
== root
->node
->start
&&
922 old_root_used
== btrfs_root_used(&root
->root_item
))
925 btrfs_set_root_node(&root
->root_item
, root
->node
);
926 ret
= btrfs_update_root(trans
, tree_root
,
932 old_root_used
= btrfs_root_used(&root
->root_item
);
933 ret
= btrfs_write_dirty_block_groups(trans
, root
);
938 if (root
!= root
->fs_info
->extent_root
)
939 switch_commit_root(root
);
945 * update all the cowonly tree roots on disk
947 * The error handling in this function may not be obvious. Any of the
948 * failures will cause the file system to go offline. We still need
949 * to clean up the delayed refs.
951 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
,
952 struct btrfs_root
*root
)
954 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
955 struct list_head
*next
;
956 struct extent_buffer
*eb
;
959 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
963 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
964 ret
= btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
,
966 btrfs_tree_unlock(eb
);
967 free_extent_buffer(eb
);
972 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
976 ret
= btrfs_run_dev_stats(trans
, root
->fs_info
);
979 ret
= btrfs_run_dev_replace(trans
, root
->fs_info
);
982 ret
= btrfs_run_qgroups(trans
, root
->fs_info
);
986 /* run_qgroups might have added some more refs */
987 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
991 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
992 next
= fs_info
->dirty_cowonly_roots
.next
;
994 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
996 ret
= update_cowonly_root(trans
, root
);
1001 down_write(&fs_info
->extent_commit_sem
);
1002 switch_commit_root(fs_info
->extent_root
);
1003 up_write(&fs_info
->extent_commit_sem
);
1005 btrfs_after_dev_replace_commit(fs_info
);
1011 * dead roots are old snapshots that need to be deleted. This allocates
1012 * a dirty root struct and adds it into the list of dead roots that need to
1015 void btrfs_add_dead_root(struct btrfs_root
*root
)
1017 spin_lock(&root
->fs_info
->trans_lock
);
1018 if (list_empty(&root
->root_list
))
1019 list_add_tail(&root
->root_list
, &root
->fs_info
->dead_roots
);
1020 spin_unlock(&root
->fs_info
->trans_lock
);
1024 * update all the cowonly tree roots on disk
1026 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
,
1027 struct btrfs_root
*root
)
1029 struct btrfs_root
*gang
[8];
1030 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1035 spin_lock(&fs_info
->fs_roots_radix_lock
);
1037 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
1040 BTRFS_ROOT_TRANS_TAG
);
1043 for (i
= 0; i
< ret
; i
++) {
1045 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
1046 (unsigned long)root
->root_key
.objectid
,
1047 BTRFS_ROOT_TRANS_TAG
);
1048 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1050 btrfs_free_log(trans
, root
);
1051 btrfs_update_reloc_root(trans
, root
);
1052 btrfs_orphan_commit_root(trans
, root
);
1054 btrfs_save_ino_cache(root
, trans
);
1056 /* see comments in should_cow_block() */
1057 root
->force_cow
= 0;
1060 if (root
->commit_root
!= root
->node
) {
1061 mutex_lock(&root
->fs_commit_mutex
);
1062 switch_commit_root(root
);
1063 btrfs_unpin_free_ino(root
);
1064 mutex_unlock(&root
->fs_commit_mutex
);
1066 btrfs_set_root_node(&root
->root_item
,
1070 err
= btrfs_update_root(trans
, fs_info
->tree_root
,
1073 spin_lock(&fs_info
->fs_roots_radix_lock
);
1078 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1083 * defrag a given btree.
1084 * Every leaf in the btree is read and defragged.
1086 int btrfs_defrag_root(struct btrfs_root
*root
)
1088 struct btrfs_fs_info
*info
= root
->fs_info
;
1089 struct btrfs_trans_handle
*trans
;
1092 if (xchg(&root
->defrag_running
, 1))
1096 trans
= btrfs_start_transaction(root
, 0);
1098 return PTR_ERR(trans
);
1100 ret
= btrfs_defrag_leaves(trans
, root
);
1102 btrfs_end_transaction(trans
, root
);
1103 btrfs_btree_balance_dirty(info
->tree_root
);
1106 if (btrfs_fs_closing(root
->fs_info
) || ret
!= -EAGAIN
)
1109 if (btrfs_defrag_cancelled(root
->fs_info
)) {
1110 pr_debug("BTRFS: defrag_root cancelled\n");
1115 root
->defrag_running
= 0;
1120 * new snapshots need to be created at a very specific time in the
1121 * transaction commit. This does the actual creation.
1124 * If the error which may affect the commitment of the current transaction
1125 * happens, we should return the error number. If the error which just affect
1126 * the creation of the pending snapshots, just return 0.
1128 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
1129 struct btrfs_fs_info
*fs_info
,
1130 struct btrfs_pending_snapshot
*pending
)
1132 struct btrfs_key key
;
1133 struct btrfs_root_item
*new_root_item
;
1134 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1135 struct btrfs_root
*root
= pending
->root
;
1136 struct btrfs_root
*parent_root
;
1137 struct btrfs_block_rsv
*rsv
;
1138 struct inode
*parent_inode
;
1139 struct btrfs_path
*path
;
1140 struct btrfs_dir_item
*dir_item
;
1141 struct dentry
*dentry
;
1142 struct extent_buffer
*tmp
;
1143 struct extent_buffer
*old
;
1144 struct timespec cur_time
= CURRENT_TIME
;
1152 path
= btrfs_alloc_path();
1154 pending
->error
= -ENOMEM
;
1158 new_root_item
= kmalloc(sizeof(*new_root_item
), GFP_NOFS
);
1159 if (!new_root_item
) {
1160 pending
->error
= -ENOMEM
;
1161 goto root_item_alloc_fail
;
1164 pending
->error
= btrfs_find_free_objectid(tree_root
, &objectid
);
1166 goto no_free_objectid
;
1168 btrfs_reloc_pre_snapshot(trans
, pending
, &to_reserve
);
1170 if (to_reserve
> 0) {
1171 pending
->error
= btrfs_block_rsv_add(root
,
1172 &pending
->block_rsv
,
1174 BTRFS_RESERVE_NO_FLUSH
);
1176 goto no_free_objectid
;
1179 pending
->error
= btrfs_qgroup_inherit(trans
, fs_info
,
1180 root
->root_key
.objectid
,
1181 objectid
, pending
->inherit
);
1183 goto no_free_objectid
;
1185 key
.objectid
= objectid
;
1186 key
.offset
= (u64
)-1;
1187 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1189 rsv
= trans
->block_rsv
;
1190 trans
->block_rsv
= &pending
->block_rsv
;
1191 trans
->bytes_reserved
= trans
->block_rsv
->reserved
;
1193 dentry
= pending
->dentry
;
1194 parent_inode
= pending
->dir
;
1195 parent_root
= BTRFS_I(parent_inode
)->root
;
1196 record_root_in_trans(trans
, parent_root
);
1199 * insert the directory item
1201 ret
= btrfs_set_inode_index(parent_inode
, &index
);
1202 BUG_ON(ret
); /* -ENOMEM */
1204 /* check if there is a file/dir which has the same name. */
1205 dir_item
= btrfs_lookup_dir_item(NULL
, parent_root
, path
,
1206 btrfs_ino(parent_inode
),
1207 dentry
->d_name
.name
,
1208 dentry
->d_name
.len
, 0);
1209 if (dir_item
!= NULL
&& !IS_ERR(dir_item
)) {
1210 pending
->error
= -EEXIST
;
1211 goto dir_item_existed
;
1212 } else if (IS_ERR(dir_item
)) {
1213 ret
= PTR_ERR(dir_item
);
1214 btrfs_abort_transaction(trans
, root
, ret
);
1217 btrfs_release_path(path
);
1220 * pull in the delayed directory update
1221 * and the delayed inode item
1222 * otherwise we corrupt the FS during
1225 ret
= btrfs_run_delayed_items(trans
, root
);
1226 if (ret
) { /* Transaction aborted */
1227 btrfs_abort_transaction(trans
, root
, ret
);
1231 record_root_in_trans(trans
, root
);
1232 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
1233 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
1234 btrfs_check_and_init_root_item(new_root_item
);
1236 root_flags
= btrfs_root_flags(new_root_item
);
1237 if (pending
->readonly
)
1238 root_flags
|= BTRFS_ROOT_SUBVOL_RDONLY
;
1240 root_flags
&= ~BTRFS_ROOT_SUBVOL_RDONLY
;
1241 btrfs_set_root_flags(new_root_item
, root_flags
);
1243 btrfs_set_root_generation_v2(new_root_item
,
1245 uuid_le_gen(&new_uuid
);
1246 memcpy(new_root_item
->uuid
, new_uuid
.b
, BTRFS_UUID_SIZE
);
1247 memcpy(new_root_item
->parent_uuid
, root
->root_item
.uuid
,
1249 if (!(root_flags
& BTRFS_ROOT_SUBVOL_RDONLY
)) {
1250 memset(new_root_item
->received_uuid
, 0,
1251 sizeof(new_root_item
->received_uuid
));
1252 memset(&new_root_item
->stime
, 0, sizeof(new_root_item
->stime
));
1253 memset(&new_root_item
->rtime
, 0, sizeof(new_root_item
->rtime
));
1254 btrfs_set_root_stransid(new_root_item
, 0);
1255 btrfs_set_root_rtransid(new_root_item
, 0);
1257 btrfs_set_stack_timespec_sec(&new_root_item
->otime
, cur_time
.tv_sec
);
1258 btrfs_set_stack_timespec_nsec(&new_root_item
->otime
, cur_time
.tv_nsec
);
1259 btrfs_set_root_otransid(new_root_item
, trans
->transid
);
1261 old
= btrfs_lock_root_node(root
);
1262 ret
= btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
);
1264 btrfs_tree_unlock(old
);
1265 free_extent_buffer(old
);
1266 btrfs_abort_transaction(trans
, root
, ret
);
1270 btrfs_set_lock_blocking(old
);
1272 ret
= btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
1273 /* clean up in any case */
1274 btrfs_tree_unlock(old
);
1275 free_extent_buffer(old
);
1277 btrfs_abort_transaction(trans
, root
, ret
);
1281 /* see comments in should_cow_block() */
1282 root
->force_cow
= 1;
1285 btrfs_set_root_node(new_root_item
, tmp
);
1286 /* record when the snapshot was created in key.offset */
1287 key
.offset
= trans
->transid
;
1288 ret
= btrfs_insert_root(trans
, tree_root
, &key
, new_root_item
);
1289 btrfs_tree_unlock(tmp
);
1290 free_extent_buffer(tmp
);
1292 btrfs_abort_transaction(trans
, root
, ret
);
1297 * insert root back/forward references
1299 ret
= btrfs_add_root_ref(trans
, tree_root
, objectid
,
1300 parent_root
->root_key
.objectid
,
1301 btrfs_ino(parent_inode
), index
,
1302 dentry
->d_name
.name
, dentry
->d_name
.len
);
1304 btrfs_abort_transaction(trans
, root
, ret
);
1308 key
.offset
= (u64
)-1;
1309 pending
->snap
= btrfs_read_fs_root_no_name(root
->fs_info
, &key
);
1310 if (IS_ERR(pending
->snap
)) {
1311 ret
= PTR_ERR(pending
->snap
);
1312 btrfs_abort_transaction(trans
, root
, ret
);
1316 ret
= btrfs_reloc_post_snapshot(trans
, pending
);
1318 btrfs_abort_transaction(trans
, root
, ret
);
1322 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1324 btrfs_abort_transaction(trans
, root
, ret
);
1328 ret
= btrfs_insert_dir_item(trans
, parent_root
,
1329 dentry
->d_name
.name
, dentry
->d_name
.len
,
1331 BTRFS_FT_DIR
, index
);
1332 /* We have check then name at the beginning, so it is impossible. */
1333 BUG_ON(ret
== -EEXIST
|| ret
== -EOVERFLOW
);
1335 btrfs_abort_transaction(trans
, root
, ret
);
1339 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
1340 dentry
->d_name
.len
* 2);
1341 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
1342 ret
= btrfs_update_inode_fallback(trans
, parent_root
, parent_inode
);
1344 btrfs_abort_transaction(trans
, root
, ret
);
1347 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
, new_uuid
.b
,
1348 BTRFS_UUID_KEY_SUBVOL
, objectid
);
1350 btrfs_abort_transaction(trans
, root
, ret
);
1353 if (!btrfs_is_empty_uuid(new_root_item
->received_uuid
)) {
1354 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
1355 new_root_item
->received_uuid
,
1356 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
1358 if (ret
&& ret
!= -EEXIST
) {
1359 btrfs_abort_transaction(trans
, root
, ret
);
1364 pending
->error
= ret
;
1366 trans
->block_rsv
= rsv
;
1367 trans
->bytes_reserved
= 0;
1369 kfree(new_root_item
);
1370 root_item_alloc_fail
:
1371 btrfs_free_path(path
);
1376 * create all the snapshots we've scheduled for creation
1378 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
,
1379 struct btrfs_fs_info
*fs_info
)
1381 struct btrfs_pending_snapshot
*pending
, *next
;
1382 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
1385 list_for_each_entry_safe(pending
, next
, head
, list
) {
1386 list_del(&pending
->list
);
1387 ret
= create_pending_snapshot(trans
, fs_info
, pending
);
1394 static void update_super_roots(struct btrfs_root
*root
)
1396 struct btrfs_root_item
*root_item
;
1397 struct btrfs_super_block
*super
;
1399 super
= root
->fs_info
->super_copy
;
1401 root_item
= &root
->fs_info
->chunk_root
->root_item
;
1402 super
->chunk_root
= root_item
->bytenr
;
1403 super
->chunk_root_generation
= root_item
->generation
;
1404 super
->chunk_root_level
= root_item
->level
;
1406 root_item
= &root
->fs_info
->tree_root
->root_item
;
1407 super
->root
= root_item
->bytenr
;
1408 super
->generation
= root_item
->generation
;
1409 super
->root_level
= root_item
->level
;
1410 if (btrfs_test_opt(root
, SPACE_CACHE
))
1411 super
->cache_generation
= root_item
->generation
;
1412 if (root
->fs_info
->update_uuid_tree_gen
)
1413 super
->uuid_tree_generation
= root_item
->generation
;
1416 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
1418 struct btrfs_transaction
*trans
;
1421 spin_lock(&info
->trans_lock
);
1422 trans
= info
->running_transaction
;
1424 ret
= (trans
->state
>= TRANS_STATE_COMMIT_START
);
1425 spin_unlock(&info
->trans_lock
);
1429 int btrfs_transaction_blocked(struct btrfs_fs_info
*info
)
1431 struct btrfs_transaction
*trans
;
1434 spin_lock(&info
->trans_lock
);
1435 trans
= info
->running_transaction
;
1437 ret
= is_transaction_blocked(trans
);
1438 spin_unlock(&info
->trans_lock
);
1443 * wait for the current transaction commit to start and block subsequent
1446 static void wait_current_trans_commit_start(struct btrfs_root
*root
,
1447 struct btrfs_transaction
*trans
)
1449 wait_event(root
->fs_info
->transaction_blocked_wait
,
1450 trans
->state
>= TRANS_STATE_COMMIT_START
||
1455 * wait for the current transaction to start and then become unblocked.
1458 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root
*root
,
1459 struct btrfs_transaction
*trans
)
1461 wait_event(root
->fs_info
->transaction_wait
,
1462 trans
->state
>= TRANS_STATE_UNBLOCKED
||
1467 * commit transactions asynchronously. once btrfs_commit_transaction_async
1468 * returns, any subsequent transaction will not be allowed to join.
1470 struct btrfs_async_commit
{
1471 struct btrfs_trans_handle
*newtrans
;
1472 struct btrfs_root
*root
;
1473 struct work_struct work
;
1476 static void do_async_commit(struct work_struct
*work
)
1478 struct btrfs_async_commit
*ac
=
1479 container_of(work
, struct btrfs_async_commit
, work
);
1482 * We've got freeze protection passed with the transaction.
1483 * Tell lockdep about it.
1485 if (ac
->newtrans
->type
& __TRANS_FREEZABLE
)
1487 &ac
->root
->fs_info
->sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
1490 current
->journal_info
= ac
->newtrans
;
1492 btrfs_commit_transaction(ac
->newtrans
, ac
->root
);
1496 int btrfs_commit_transaction_async(struct btrfs_trans_handle
*trans
,
1497 struct btrfs_root
*root
,
1498 int wait_for_unblock
)
1500 struct btrfs_async_commit
*ac
;
1501 struct btrfs_transaction
*cur_trans
;
1503 ac
= kmalloc(sizeof(*ac
), GFP_NOFS
);
1507 INIT_WORK(&ac
->work
, do_async_commit
);
1509 ac
->newtrans
= btrfs_join_transaction(root
);
1510 if (IS_ERR(ac
->newtrans
)) {
1511 int err
= PTR_ERR(ac
->newtrans
);
1516 /* take transaction reference */
1517 cur_trans
= trans
->transaction
;
1518 atomic_inc(&cur_trans
->use_count
);
1520 btrfs_end_transaction(trans
, root
);
1523 * Tell lockdep we've released the freeze rwsem, since the
1524 * async commit thread will be the one to unlock it.
1526 if (ac
->newtrans
->type
& __TRANS_FREEZABLE
)
1528 &root
->fs_info
->sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
1531 schedule_work(&ac
->work
);
1533 /* wait for transaction to start and unblock */
1534 if (wait_for_unblock
)
1535 wait_current_trans_commit_start_and_unblock(root
, cur_trans
);
1537 wait_current_trans_commit_start(root
, cur_trans
);
1539 if (current
->journal_info
== trans
)
1540 current
->journal_info
= NULL
;
1542 btrfs_put_transaction(cur_trans
);
1547 static void cleanup_transaction(struct btrfs_trans_handle
*trans
,
1548 struct btrfs_root
*root
, int err
)
1550 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1553 WARN_ON(trans
->use_count
> 1);
1555 btrfs_abort_transaction(trans
, root
, err
);
1557 spin_lock(&root
->fs_info
->trans_lock
);
1560 * If the transaction is removed from the list, it means this
1561 * transaction has been committed successfully, so it is impossible
1562 * to call the cleanup function.
1564 BUG_ON(list_empty(&cur_trans
->list
));
1566 list_del_init(&cur_trans
->list
);
1567 if (cur_trans
== root
->fs_info
->running_transaction
) {
1568 cur_trans
->state
= TRANS_STATE_COMMIT_DOING
;
1569 spin_unlock(&root
->fs_info
->trans_lock
);
1570 wait_event(cur_trans
->writer_wait
,
1571 atomic_read(&cur_trans
->num_writers
) == 1);
1573 spin_lock(&root
->fs_info
->trans_lock
);
1575 spin_unlock(&root
->fs_info
->trans_lock
);
1577 btrfs_cleanup_one_transaction(trans
->transaction
, root
);
1579 spin_lock(&root
->fs_info
->trans_lock
);
1580 if (cur_trans
== root
->fs_info
->running_transaction
)
1581 root
->fs_info
->running_transaction
= NULL
;
1582 spin_unlock(&root
->fs_info
->trans_lock
);
1584 if (trans
->type
& __TRANS_FREEZABLE
)
1585 sb_end_intwrite(root
->fs_info
->sb
);
1586 btrfs_put_transaction(cur_trans
);
1587 btrfs_put_transaction(cur_trans
);
1589 trace_btrfs_transaction_commit(root
);
1591 btrfs_scrub_continue(root
);
1593 if (current
->journal_info
== trans
)
1594 current
->journal_info
= NULL
;
1596 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1599 static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle
*trans
,
1600 struct btrfs_root
*root
)
1604 ret
= btrfs_run_delayed_items(trans
, root
);
1606 * running the delayed items may have added new refs. account
1607 * them now so that they hinder processing of more delayed refs
1608 * as little as possible.
1611 btrfs_delayed_refs_qgroup_accounting(trans
, root
->fs_info
);
1615 ret
= btrfs_delayed_refs_qgroup_accounting(trans
, root
->fs_info
);
1620 * rename don't use btrfs_join_transaction, so, once we
1621 * set the transaction to blocked above, we aren't going
1622 * to get any new ordered operations. We can safely run
1623 * it here and no for sure that nothing new will be added
1626 ret
= btrfs_run_ordered_operations(trans
, root
, 1);
1631 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info
*fs_info
)
1633 if (btrfs_test_opt(fs_info
->tree_root
, FLUSHONCOMMIT
))
1634 return btrfs_start_delalloc_roots(fs_info
, 1);
1638 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info
*fs_info
)
1640 if (btrfs_test_opt(fs_info
->tree_root
, FLUSHONCOMMIT
))
1641 btrfs_wait_ordered_roots(fs_info
, -1);
1644 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
,
1645 struct btrfs_root
*root
)
1647 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1648 struct btrfs_transaction
*prev_trans
= NULL
;
1651 ret
= btrfs_run_ordered_operations(trans
, root
, 0);
1653 btrfs_abort_transaction(trans
, root
, ret
);
1654 btrfs_end_transaction(trans
, root
);
1658 /* Stop the commit early if ->aborted is set */
1659 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1660 ret
= cur_trans
->aborted
;
1661 btrfs_end_transaction(trans
, root
);
1665 /* make a pass through all the delayed refs we have so far
1666 * any runnings procs may add more while we are here
1668 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1670 btrfs_end_transaction(trans
, root
);
1674 btrfs_trans_release_metadata(trans
, root
);
1675 trans
->block_rsv
= NULL
;
1676 if (trans
->qgroup_reserved
) {
1677 btrfs_qgroup_free(root
, trans
->qgroup_reserved
);
1678 trans
->qgroup_reserved
= 0;
1681 cur_trans
= trans
->transaction
;
1684 * set the flushing flag so procs in this transaction have to
1685 * start sending their work down.
1687 cur_trans
->delayed_refs
.flushing
= 1;
1690 if (!list_empty(&trans
->new_bgs
))
1691 btrfs_create_pending_block_groups(trans
, root
);
1693 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1695 btrfs_end_transaction(trans
, root
);
1699 spin_lock(&root
->fs_info
->trans_lock
);
1700 if (cur_trans
->state
>= TRANS_STATE_COMMIT_START
) {
1701 spin_unlock(&root
->fs_info
->trans_lock
);
1702 atomic_inc(&cur_trans
->use_count
);
1703 ret
= btrfs_end_transaction(trans
, root
);
1705 wait_for_commit(root
, cur_trans
);
1707 btrfs_put_transaction(cur_trans
);
1712 cur_trans
->state
= TRANS_STATE_COMMIT_START
;
1713 wake_up(&root
->fs_info
->transaction_blocked_wait
);
1715 if (cur_trans
->list
.prev
!= &root
->fs_info
->trans_list
) {
1716 prev_trans
= list_entry(cur_trans
->list
.prev
,
1717 struct btrfs_transaction
, list
);
1718 if (prev_trans
->state
!= TRANS_STATE_COMPLETED
) {
1719 atomic_inc(&prev_trans
->use_count
);
1720 spin_unlock(&root
->fs_info
->trans_lock
);
1722 wait_for_commit(root
, prev_trans
);
1724 btrfs_put_transaction(prev_trans
);
1726 spin_unlock(&root
->fs_info
->trans_lock
);
1729 spin_unlock(&root
->fs_info
->trans_lock
);
1732 extwriter_counter_dec(cur_trans
, trans
->type
);
1734 ret
= btrfs_start_delalloc_flush(root
->fs_info
);
1736 goto cleanup_transaction
;
1738 ret
= btrfs_flush_all_pending_stuffs(trans
, root
);
1740 goto cleanup_transaction
;
1742 wait_event(cur_trans
->writer_wait
,
1743 extwriter_counter_read(cur_trans
) == 0);
1745 /* some pending stuffs might be added after the previous flush. */
1746 ret
= btrfs_flush_all_pending_stuffs(trans
, root
);
1748 goto cleanup_transaction
;
1750 btrfs_wait_delalloc_flush(root
->fs_info
);
1752 btrfs_scrub_pause(root
);
1754 * Ok now we need to make sure to block out any other joins while we
1755 * commit the transaction. We could have started a join before setting
1756 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
1758 spin_lock(&root
->fs_info
->trans_lock
);
1759 cur_trans
->state
= TRANS_STATE_COMMIT_DOING
;
1760 spin_unlock(&root
->fs_info
->trans_lock
);
1761 wait_event(cur_trans
->writer_wait
,
1762 atomic_read(&cur_trans
->num_writers
) == 1);
1764 /* ->aborted might be set after the previous check, so check it */
1765 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1766 ret
= cur_trans
->aborted
;
1767 goto cleanup_transaction
;
1770 * the reloc mutex makes sure that we stop
1771 * the balancing code from coming in and moving
1772 * extents around in the middle of the commit
1774 mutex_lock(&root
->fs_info
->reloc_mutex
);
1777 * We needn't worry about the delayed items because we will
1778 * deal with them in create_pending_snapshot(), which is the
1779 * core function of the snapshot creation.
1781 ret
= create_pending_snapshots(trans
, root
->fs_info
);
1783 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1784 goto cleanup_transaction
;
1788 * We insert the dir indexes of the snapshots and update the inode
1789 * of the snapshots' parents after the snapshot creation, so there
1790 * are some delayed items which are not dealt with. Now deal with
1793 * We needn't worry that this operation will corrupt the snapshots,
1794 * because all the tree which are snapshoted will be forced to COW
1795 * the nodes and leaves.
1797 ret
= btrfs_run_delayed_items(trans
, root
);
1799 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1800 goto cleanup_transaction
;
1803 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1805 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1806 goto cleanup_transaction
;
1810 * make sure none of the code above managed to slip in a
1813 btrfs_assert_delayed_root_empty(root
);
1815 WARN_ON(cur_trans
!= trans
->transaction
);
1817 /* btrfs_commit_tree_roots is responsible for getting the
1818 * various roots consistent with each other. Every pointer
1819 * in the tree of tree roots has to point to the most up to date
1820 * root for every subvolume and other tree. So, we have to keep
1821 * the tree logging code from jumping in and changing any
1824 * At this point in the commit, there can't be any tree-log
1825 * writers, but a little lower down we drop the trans mutex
1826 * and let new people in. By holding the tree_log_mutex
1827 * from now until after the super is written, we avoid races
1828 * with the tree-log code.
1830 mutex_lock(&root
->fs_info
->tree_log_mutex
);
1832 ret
= commit_fs_roots(trans
, root
);
1834 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1835 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1836 goto cleanup_transaction
;
1839 /* commit_fs_roots gets rid of all the tree log roots, it is now
1840 * safe to free the root of tree log roots
1842 btrfs_free_log_root_tree(trans
, root
->fs_info
);
1844 ret
= commit_cowonly_roots(trans
, root
);
1846 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1847 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1848 goto cleanup_transaction
;
1852 * The tasks which save the space cache and inode cache may also
1853 * update ->aborted, check it.
1855 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1856 ret
= cur_trans
->aborted
;
1857 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1858 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1859 goto cleanup_transaction
;
1862 btrfs_prepare_extent_commit(trans
, root
);
1864 cur_trans
= root
->fs_info
->running_transaction
;
1866 btrfs_set_root_node(&root
->fs_info
->tree_root
->root_item
,
1867 root
->fs_info
->tree_root
->node
);
1868 switch_commit_root(root
->fs_info
->tree_root
);
1870 btrfs_set_root_node(&root
->fs_info
->chunk_root
->root_item
,
1871 root
->fs_info
->chunk_root
->node
);
1872 switch_commit_root(root
->fs_info
->chunk_root
);
1874 assert_qgroups_uptodate(trans
);
1875 update_super_roots(root
);
1877 btrfs_set_super_log_root(root
->fs_info
->super_copy
, 0);
1878 btrfs_set_super_log_root_level(root
->fs_info
->super_copy
, 0);
1879 memcpy(root
->fs_info
->super_for_commit
, root
->fs_info
->super_copy
,
1880 sizeof(*root
->fs_info
->super_copy
));
1882 spin_lock(&root
->fs_info
->trans_lock
);
1883 cur_trans
->state
= TRANS_STATE_UNBLOCKED
;
1884 root
->fs_info
->running_transaction
= NULL
;
1885 spin_unlock(&root
->fs_info
->trans_lock
);
1886 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1888 wake_up(&root
->fs_info
->transaction_wait
);
1890 ret
= btrfs_write_and_wait_transaction(trans
, root
);
1892 btrfs_error(root
->fs_info
, ret
,
1893 "Error while writing out transaction");
1894 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1895 goto cleanup_transaction
;
1898 ret
= write_ctree_super(trans
, root
, 0);
1900 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1901 goto cleanup_transaction
;
1905 * the super is written, we can safely allow the tree-loggers
1906 * to go about their business
1908 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1910 btrfs_finish_extent_commit(trans
, root
);
1912 root
->fs_info
->last_trans_committed
= cur_trans
->transid
;
1914 * We needn't acquire the lock here because there is no other task
1915 * which can change it.
1917 cur_trans
->state
= TRANS_STATE_COMPLETED
;
1918 wake_up(&cur_trans
->commit_wait
);
1920 spin_lock(&root
->fs_info
->trans_lock
);
1921 list_del_init(&cur_trans
->list
);
1922 spin_unlock(&root
->fs_info
->trans_lock
);
1924 btrfs_put_transaction(cur_trans
);
1925 btrfs_put_transaction(cur_trans
);
1927 if (trans
->type
& __TRANS_FREEZABLE
)
1928 sb_end_intwrite(root
->fs_info
->sb
);
1930 trace_btrfs_transaction_commit(root
);
1932 btrfs_scrub_continue(root
);
1934 if (current
->journal_info
== trans
)
1935 current
->journal_info
= NULL
;
1937 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1939 if (current
!= root
->fs_info
->transaction_kthread
)
1940 btrfs_run_delayed_iputs(root
);
1944 cleanup_transaction
:
1945 btrfs_trans_release_metadata(trans
, root
);
1946 trans
->block_rsv
= NULL
;
1947 if (trans
->qgroup_reserved
) {
1948 btrfs_qgroup_free(root
, trans
->qgroup_reserved
);
1949 trans
->qgroup_reserved
= 0;
1951 btrfs_warn(root
->fs_info
, "Skipping commit of aborted transaction.");
1952 if (current
->journal_info
== trans
)
1953 current
->journal_info
= NULL
;
1954 cleanup_transaction(trans
, root
, ret
);
1960 * return < 0 if error
1961 * 0 if there are no more dead_roots at the time of call
1962 * 1 there are more to be processed, call me again
1964 * The return value indicates there are certainly more snapshots to delete, but
1965 * if there comes a new one during processing, it may return 0. We don't mind,
1966 * because btrfs_commit_super will poke cleaner thread and it will process it a
1967 * few seconds later.
1969 int btrfs_clean_one_deleted_snapshot(struct btrfs_root
*root
)
1972 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1974 spin_lock(&fs_info
->trans_lock
);
1975 if (list_empty(&fs_info
->dead_roots
)) {
1976 spin_unlock(&fs_info
->trans_lock
);
1979 root
= list_first_entry(&fs_info
->dead_roots
,
1980 struct btrfs_root
, root_list
);
1981 list_del_init(&root
->root_list
);
1982 spin_unlock(&fs_info
->trans_lock
);
1984 pr_debug("BTRFS: cleaner removing %llu\n", root
->objectid
);
1986 btrfs_kill_all_delayed_nodes(root
);
1988 if (btrfs_header_backref_rev(root
->node
) <
1989 BTRFS_MIXED_BACKREF_REV
)
1990 ret
= btrfs_drop_snapshot(root
, NULL
, 0, 0);
1992 ret
= btrfs_drop_snapshot(root
, NULL
, 1, 0);
1994 * If we encounter a transaction abort during snapshot cleaning, we
1995 * don't want to crash here
1997 return (ret
< 0) ? 0 : 1;