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
> 1) {
690 trans
->block_rsv
= trans
->orig_rsv
;
695 * do the qgroup accounting as early as possible
697 err
= btrfs_delayed_refs_qgroup_accounting(trans
, info
);
699 btrfs_trans_release_metadata(trans
, root
);
700 trans
->block_rsv
= NULL
;
702 if (trans
->qgroup_reserved
) {
704 * the same root has to be passed here between start_transaction
705 * and end_transaction. Subvolume quota depends on this.
707 btrfs_qgroup_free(trans
->root
, trans
->qgroup_reserved
);
708 trans
->qgroup_reserved
= 0;
711 if (!list_empty(&trans
->new_bgs
))
712 btrfs_create_pending_block_groups(trans
, root
);
714 trans
->delayed_ref_updates
= 0;
715 if (btrfs_should_throttle_delayed_refs(trans
, root
)) {
716 cur
= max_t(unsigned long, cur
, 1);
717 trans
->delayed_ref_updates
= 0;
718 btrfs_run_delayed_refs(trans
, root
, cur
);
721 btrfs_trans_release_metadata(trans
, root
);
722 trans
->block_rsv
= NULL
;
724 if (!list_empty(&trans
->new_bgs
))
725 btrfs_create_pending_block_groups(trans
, root
);
727 if (lock
&& !atomic_read(&root
->fs_info
->open_ioctl_trans
) &&
728 should_end_transaction(trans
, root
) &&
729 ACCESS_ONCE(cur_trans
->state
) == TRANS_STATE_RUNNING
) {
730 spin_lock(&info
->trans_lock
);
731 if (cur_trans
->state
== TRANS_STATE_RUNNING
)
732 cur_trans
->state
= TRANS_STATE_BLOCKED
;
733 spin_unlock(&info
->trans_lock
);
736 if (lock
&& ACCESS_ONCE(cur_trans
->state
) == TRANS_STATE_BLOCKED
) {
738 return btrfs_commit_transaction(trans
, root
);
740 wake_up_process(info
->transaction_kthread
);
743 if (trans
->type
& __TRANS_FREEZABLE
)
744 sb_end_intwrite(root
->fs_info
->sb
);
746 WARN_ON(cur_trans
!= info
->running_transaction
);
747 WARN_ON(atomic_read(&cur_trans
->num_writers
) < 1);
748 atomic_dec(&cur_trans
->num_writers
);
749 extwriter_counter_dec(cur_trans
, trans
->type
);
752 if (waitqueue_active(&cur_trans
->writer_wait
))
753 wake_up(&cur_trans
->writer_wait
);
754 btrfs_put_transaction(cur_trans
);
756 if (current
->journal_info
== trans
)
757 current
->journal_info
= NULL
;
760 btrfs_run_delayed_iputs(root
);
762 if (trans
->aborted
||
763 test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
)) {
764 wake_up_process(info
->transaction_kthread
);
767 assert_qgroups_uptodate(trans
);
769 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
773 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
774 struct btrfs_root
*root
)
776 return __btrfs_end_transaction(trans
, root
, 0);
779 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
,
780 struct btrfs_root
*root
)
782 return __btrfs_end_transaction(trans
, root
, 1);
785 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle
*trans
,
786 struct btrfs_root
*root
)
788 return __btrfs_end_transaction(trans
, root
, 1);
792 * when btree blocks are allocated, they have some corresponding bits set for
793 * them in one of two extent_io trees. This is used to make sure all of
794 * those extents are sent to disk but does not wait on them
796 int btrfs_write_marked_extents(struct btrfs_root
*root
,
797 struct extent_io_tree
*dirty_pages
, int mark
)
801 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
802 struct extent_state
*cached_state
= NULL
;
806 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
807 mark
, &cached_state
)) {
808 convert_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
,
809 mark
, &cached_state
, GFP_NOFS
);
811 err
= filemap_fdatawrite_range(mapping
, start
, end
);
823 * when btree blocks are allocated, they have some corresponding bits set for
824 * them in one of two extent_io trees. This is used to make sure all of
825 * those extents are on disk for transaction or log commit. We wait
826 * on all the pages and clear them from the dirty pages state tree
828 int btrfs_wait_marked_extents(struct btrfs_root
*root
,
829 struct extent_io_tree
*dirty_pages
, int mark
)
833 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
834 struct extent_state
*cached_state
= NULL
;
838 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
839 EXTENT_NEED_WAIT
, &cached_state
)) {
840 clear_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
,
841 0, 0, &cached_state
, GFP_NOFS
);
842 err
= filemap_fdatawait_range(mapping
, start
, end
);
854 * when btree blocks are allocated, they have some corresponding bits set for
855 * them in one of two extent_io trees. This is used to make sure all of
856 * those extents are on disk for transaction or log commit
858 static int btrfs_write_and_wait_marked_extents(struct btrfs_root
*root
,
859 struct extent_io_tree
*dirty_pages
, int mark
)
863 struct blk_plug plug
;
865 blk_start_plug(&plug
);
866 ret
= btrfs_write_marked_extents(root
, dirty_pages
, mark
);
867 blk_finish_plug(&plug
);
868 ret2
= btrfs_wait_marked_extents(root
, dirty_pages
, mark
);
877 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
,
878 struct btrfs_root
*root
)
880 if (!trans
|| !trans
->transaction
) {
881 struct inode
*btree_inode
;
882 btree_inode
= root
->fs_info
->btree_inode
;
883 return filemap_write_and_wait(btree_inode
->i_mapping
);
885 return btrfs_write_and_wait_marked_extents(root
,
886 &trans
->transaction
->dirty_pages
,
891 * this is used to update the root pointer in the tree of tree roots.
893 * But, in the case of the extent allocation tree, updating the root
894 * pointer may allocate blocks which may change the root of the extent
897 * So, this loops and repeats and makes sure the cowonly root didn't
898 * change while the root pointer was being updated in the metadata.
900 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
901 struct btrfs_root
*root
)
906 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
908 old_root_used
= btrfs_root_used(&root
->root_item
);
909 btrfs_write_dirty_block_groups(trans
, root
);
912 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
913 if (old_root_bytenr
== root
->node
->start
&&
914 old_root_used
== btrfs_root_used(&root
->root_item
))
917 btrfs_set_root_node(&root
->root_item
, root
->node
);
918 ret
= btrfs_update_root(trans
, tree_root
,
924 old_root_used
= btrfs_root_used(&root
->root_item
);
925 ret
= btrfs_write_dirty_block_groups(trans
, root
);
930 if (root
!= root
->fs_info
->extent_root
)
931 switch_commit_root(root
);
937 * update all the cowonly tree roots on disk
939 * The error handling in this function may not be obvious. Any of the
940 * failures will cause the file system to go offline. We still need
941 * to clean up the delayed refs.
943 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
,
944 struct btrfs_root
*root
)
946 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
947 struct list_head
*next
;
948 struct extent_buffer
*eb
;
951 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
955 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
956 ret
= btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
,
958 btrfs_tree_unlock(eb
);
959 free_extent_buffer(eb
);
964 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
968 ret
= btrfs_run_dev_stats(trans
, root
->fs_info
);
970 ret
= btrfs_run_dev_replace(trans
, root
->fs_info
);
973 ret
= btrfs_run_qgroups(trans
, root
->fs_info
);
976 /* run_qgroups might have added some more refs */
977 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
980 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
981 next
= fs_info
->dirty_cowonly_roots
.next
;
983 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
985 ret
= update_cowonly_root(trans
, root
);
990 down_write(&fs_info
->extent_commit_sem
);
991 switch_commit_root(fs_info
->extent_root
);
992 up_write(&fs_info
->extent_commit_sem
);
994 btrfs_after_dev_replace_commit(fs_info
);
1000 * dead roots are old snapshots that need to be deleted. This allocates
1001 * a dirty root struct and adds it into the list of dead roots that need to
1004 void btrfs_add_dead_root(struct btrfs_root
*root
)
1006 spin_lock(&root
->fs_info
->trans_lock
);
1007 if (list_empty(&root
->root_list
))
1008 list_add_tail(&root
->root_list
, &root
->fs_info
->dead_roots
);
1009 spin_unlock(&root
->fs_info
->trans_lock
);
1013 * update all the cowonly tree roots on disk
1015 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
,
1016 struct btrfs_root
*root
)
1018 struct btrfs_root
*gang
[8];
1019 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1024 spin_lock(&fs_info
->fs_roots_radix_lock
);
1026 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
1029 BTRFS_ROOT_TRANS_TAG
);
1032 for (i
= 0; i
< ret
; i
++) {
1034 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
1035 (unsigned long)root
->root_key
.objectid
,
1036 BTRFS_ROOT_TRANS_TAG
);
1037 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1039 btrfs_free_log(trans
, root
);
1040 btrfs_update_reloc_root(trans
, root
);
1041 btrfs_orphan_commit_root(trans
, root
);
1043 btrfs_save_ino_cache(root
, trans
);
1045 /* see comments in should_cow_block() */
1046 root
->force_cow
= 0;
1049 if (root
->commit_root
!= root
->node
) {
1050 mutex_lock(&root
->fs_commit_mutex
);
1051 switch_commit_root(root
);
1052 btrfs_unpin_free_ino(root
);
1053 mutex_unlock(&root
->fs_commit_mutex
);
1055 btrfs_set_root_node(&root
->root_item
,
1059 err
= btrfs_update_root(trans
, fs_info
->tree_root
,
1062 spin_lock(&fs_info
->fs_roots_radix_lock
);
1067 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1072 * defrag a given btree.
1073 * Every leaf in the btree is read and defragged.
1075 int btrfs_defrag_root(struct btrfs_root
*root
)
1077 struct btrfs_fs_info
*info
= root
->fs_info
;
1078 struct btrfs_trans_handle
*trans
;
1081 if (xchg(&root
->defrag_running
, 1))
1085 trans
= btrfs_start_transaction(root
, 0);
1087 return PTR_ERR(trans
);
1089 ret
= btrfs_defrag_leaves(trans
, root
);
1091 btrfs_end_transaction(trans
, root
);
1092 btrfs_btree_balance_dirty(info
->tree_root
);
1095 if (btrfs_fs_closing(root
->fs_info
) || ret
!= -EAGAIN
)
1098 if (btrfs_defrag_cancelled(root
->fs_info
)) {
1099 printk(KERN_DEBUG
"btrfs: defrag_root cancelled\n");
1104 root
->defrag_running
= 0;
1109 * new snapshots need to be created at a very specific time in the
1110 * transaction commit. This does the actual creation.
1113 * If the error which may affect the commitment of the current transaction
1114 * happens, we should return the error number. If the error which just affect
1115 * the creation of the pending snapshots, just return 0.
1117 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
1118 struct btrfs_fs_info
*fs_info
,
1119 struct btrfs_pending_snapshot
*pending
)
1121 struct btrfs_key key
;
1122 struct btrfs_root_item
*new_root_item
;
1123 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1124 struct btrfs_root
*root
= pending
->root
;
1125 struct btrfs_root
*parent_root
;
1126 struct btrfs_block_rsv
*rsv
;
1127 struct inode
*parent_inode
;
1128 struct btrfs_path
*path
;
1129 struct btrfs_dir_item
*dir_item
;
1130 struct dentry
*dentry
;
1131 struct extent_buffer
*tmp
;
1132 struct extent_buffer
*old
;
1133 struct timespec cur_time
= CURRENT_TIME
;
1141 path
= btrfs_alloc_path();
1143 pending
->error
= -ENOMEM
;
1147 new_root_item
= kmalloc(sizeof(*new_root_item
), GFP_NOFS
);
1148 if (!new_root_item
) {
1149 pending
->error
= -ENOMEM
;
1150 goto root_item_alloc_fail
;
1153 pending
->error
= btrfs_find_free_objectid(tree_root
, &objectid
);
1155 goto no_free_objectid
;
1157 btrfs_reloc_pre_snapshot(trans
, pending
, &to_reserve
);
1159 if (to_reserve
> 0) {
1160 pending
->error
= btrfs_block_rsv_add(root
,
1161 &pending
->block_rsv
,
1163 BTRFS_RESERVE_NO_FLUSH
);
1165 goto no_free_objectid
;
1168 pending
->error
= btrfs_qgroup_inherit(trans
, fs_info
,
1169 root
->root_key
.objectid
,
1170 objectid
, pending
->inherit
);
1172 goto no_free_objectid
;
1174 key
.objectid
= objectid
;
1175 key
.offset
= (u64
)-1;
1176 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1178 rsv
= trans
->block_rsv
;
1179 trans
->block_rsv
= &pending
->block_rsv
;
1180 trans
->bytes_reserved
= trans
->block_rsv
->reserved
;
1182 dentry
= pending
->dentry
;
1183 parent_inode
= pending
->dir
;
1184 parent_root
= BTRFS_I(parent_inode
)->root
;
1185 record_root_in_trans(trans
, parent_root
);
1188 * insert the directory item
1190 ret
= btrfs_set_inode_index(parent_inode
, &index
);
1191 BUG_ON(ret
); /* -ENOMEM */
1193 /* check if there is a file/dir which has the same name. */
1194 dir_item
= btrfs_lookup_dir_item(NULL
, parent_root
, path
,
1195 btrfs_ino(parent_inode
),
1196 dentry
->d_name
.name
,
1197 dentry
->d_name
.len
, 0);
1198 if (dir_item
!= NULL
&& !IS_ERR(dir_item
)) {
1199 pending
->error
= -EEXIST
;
1200 goto dir_item_existed
;
1201 } else if (IS_ERR(dir_item
)) {
1202 ret
= PTR_ERR(dir_item
);
1203 btrfs_abort_transaction(trans
, root
, ret
);
1206 btrfs_release_path(path
);
1209 * pull in the delayed directory update
1210 * and the delayed inode item
1211 * otherwise we corrupt the FS during
1214 ret
= btrfs_run_delayed_items(trans
, root
);
1215 if (ret
) { /* Transaction aborted */
1216 btrfs_abort_transaction(trans
, root
, ret
);
1220 record_root_in_trans(trans
, root
);
1221 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
1222 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
1223 btrfs_check_and_init_root_item(new_root_item
);
1225 root_flags
= btrfs_root_flags(new_root_item
);
1226 if (pending
->readonly
)
1227 root_flags
|= BTRFS_ROOT_SUBVOL_RDONLY
;
1229 root_flags
&= ~BTRFS_ROOT_SUBVOL_RDONLY
;
1230 btrfs_set_root_flags(new_root_item
, root_flags
);
1232 btrfs_set_root_generation_v2(new_root_item
,
1234 uuid_le_gen(&new_uuid
);
1235 memcpy(new_root_item
->uuid
, new_uuid
.b
, BTRFS_UUID_SIZE
);
1236 memcpy(new_root_item
->parent_uuid
, root
->root_item
.uuid
,
1238 if (!(root_flags
& BTRFS_ROOT_SUBVOL_RDONLY
)) {
1239 memset(new_root_item
->received_uuid
, 0,
1240 sizeof(new_root_item
->received_uuid
));
1241 memset(&new_root_item
->stime
, 0, sizeof(new_root_item
->stime
));
1242 memset(&new_root_item
->rtime
, 0, sizeof(new_root_item
->rtime
));
1243 btrfs_set_root_stransid(new_root_item
, 0);
1244 btrfs_set_root_rtransid(new_root_item
, 0);
1246 btrfs_set_stack_timespec_sec(&new_root_item
->otime
, cur_time
.tv_sec
);
1247 btrfs_set_stack_timespec_nsec(&new_root_item
->otime
, cur_time
.tv_nsec
);
1248 btrfs_set_root_otransid(new_root_item
, trans
->transid
);
1250 old
= btrfs_lock_root_node(root
);
1251 ret
= btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
);
1253 btrfs_tree_unlock(old
);
1254 free_extent_buffer(old
);
1255 btrfs_abort_transaction(trans
, root
, ret
);
1259 btrfs_set_lock_blocking(old
);
1261 ret
= btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
1262 /* clean up in any case */
1263 btrfs_tree_unlock(old
);
1264 free_extent_buffer(old
);
1266 btrfs_abort_transaction(trans
, root
, ret
);
1270 /* see comments in should_cow_block() */
1271 root
->force_cow
= 1;
1274 btrfs_set_root_node(new_root_item
, tmp
);
1275 /* record when the snapshot was created in key.offset */
1276 key
.offset
= trans
->transid
;
1277 ret
= btrfs_insert_root(trans
, tree_root
, &key
, new_root_item
);
1278 btrfs_tree_unlock(tmp
);
1279 free_extent_buffer(tmp
);
1281 btrfs_abort_transaction(trans
, root
, ret
);
1286 * insert root back/forward references
1288 ret
= btrfs_add_root_ref(trans
, tree_root
, objectid
,
1289 parent_root
->root_key
.objectid
,
1290 btrfs_ino(parent_inode
), index
,
1291 dentry
->d_name
.name
, dentry
->d_name
.len
);
1293 btrfs_abort_transaction(trans
, root
, ret
);
1297 key
.offset
= (u64
)-1;
1298 pending
->snap
= btrfs_read_fs_root_no_name(root
->fs_info
, &key
);
1299 if (IS_ERR(pending
->snap
)) {
1300 ret
= PTR_ERR(pending
->snap
);
1301 btrfs_abort_transaction(trans
, root
, ret
);
1305 ret
= btrfs_reloc_post_snapshot(trans
, pending
);
1307 btrfs_abort_transaction(trans
, root
, ret
);
1311 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1313 btrfs_abort_transaction(trans
, root
, ret
);
1317 ret
= btrfs_insert_dir_item(trans
, parent_root
,
1318 dentry
->d_name
.name
, dentry
->d_name
.len
,
1320 BTRFS_FT_DIR
, index
);
1321 /* We have check then name at the beginning, so it is impossible. */
1322 BUG_ON(ret
== -EEXIST
|| ret
== -EOVERFLOW
);
1324 btrfs_abort_transaction(trans
, root
, ret
);
1328 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
1329 dentry
->d_name
.len
* 2);
1330 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
1331 ret
= btrfs_update_inode_fallback(trans
, parent_root
, parent_inode
);
1333 btrfs_abort_transaction(trans
, root
, ret
);
1336 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
, new_uuid
.b
,
1337 BTRFS_UUID_KEY_SUBVOL
, objectid
);
1339 btrfs_abort_transaction(trans
, root
, ret
);
1342 if (!btrfs_is_empty_uuid(new_root_item
->received_uuid
)) {
1343 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
1344 new_root_item
->received_uuid
,
1345 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
1347 if (ret
&& ret
!= -EEXIST
) {
1348 btrfs_abort_transaction(trans
, root
, ret
);
1353 pending
->error
= ret
;
1355 trans
->block_rsv
= rsv
;
1356 trans
->bytes_reserved
= 0;
1358 kfree(new_root_item
);
1359 root_item_alloc_fail
:
1360 btrfs_free_path(path
);
1365 * create all the snapshots we've scheduled for creation
1367 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
,
1368 struct btrfs_fs_info
*fs_info
)
1370 struct btrfs_pending_snapshot
*pending
, *next
;
1371 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
1374 list_for_each_entry_safe(pending
, next
, head
, list
) {
1375 list_del(&pending
->list
);
1376 ret
= create_pending_snapshot(trans
, fs_info
, pending
);
1383 static void update_super_roots(struct btrfs_root
*root
)
1385 struct btrfs_root_item
*root_item
;
1386 struct btrfs_super_block
*super
;
1388 super
= root
->fs_info
->super_copy
;
1390 root_item
= &root
->fs_info
->chunk_root
->root_item
;
1391 super
->chunk_root
= root_item
->bytenr
;
1392 super
->chunk_root_generation
= root_item
->generation
;
1393 super
->chunk_root_level
= root_item
->level
;
1395 root_item
= &root
->fs_info
->tree_root
->root_item
;
1396 super
->root
= root_item
->bytenr
;
1397 super
->generation
= root_item
->generation
;
1398 super
->root_level
= root_item
->level
;
1399 if (btrfs_test_opt(root
, SPACE_CACHE
))
1400 super
->cache_generation
= root_item
->generation
;
1401 if (root
->fs_info
->update_uuid_tree_gen
)
1402 super
->uuid_tree_generation
= root_item
->generation
;
1405 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
1407 struct btrfs_transaction
*trans
;
1410 spin_lock(&info
->trans_lock
);
1411 trans
= info
->running_transaction
;
1413 ret
= (trans
->state
>= TRANS_STATE_COMMIT_START
);
1414 spin_unlock(&info
->trans_lock
);
1418 int btrfs_transaction_blocked(struct btrfs_fs_info
*info
)
1420 struct btrfs_transaction
*trans
;
1423 spin_lock(&info
->trans_lock
);
1424 trans
= info
->running_transaction
;
1426 ret
= is_transaction_blocked(trans
);
1427 spin_unlock(&info
->trans_lock
);
1432 * wait for the current transaction commit to start and block subsequent
1435 static void wait_current_trans_commit_start(struct btrfs_root
*root
,
1436 struct btrfs_transaction
*trans
)
1438 wait_event(root
->fs_info
->transaction_blocked_wait
,
1439 trans
->state
>= TRANS_STATE_COMMIT_START
||
1444 * wait for the current transaction to start and then become unblocked.
1447 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root
*root
,
1448 struct btrfs_transaction
*trans
)
1450 wait_event(root
->fs_info
->transaction_wait
,
1451 trans
->state
>= TRANS_STATE_UNBLOCKED
||
1456 * commit transactions asynchronously. once btrfs_commit_transaction_async
1457 * returns, any subsequent transaction will not be allowed to join.
1459 struct btrfs_async_commit
{
1460 struct btrfs_trans_handle
*newtrans
;
1461 struct btrfs_root
*root
;
1462 struct work_struct work
;
1465 static void do_async_commit(struct work_struct
*work
)
1467 struct btrfs_async_commit
*ac
=
1468 container_of(work
, struct btrfs_async_commit
, work
);
1471 * We've got freeze protection passed with the transaction.
1472 * Tell lockdep about it.
1474 if (ac
->newtrans
->type
& __TRANS_FREEZABLE
)
1476 &ac
->root
->fs_info
->sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
1479 current
->journal_info
= ac
->newtrans
;
1481 btrfs_commit_transaction(ac
->newtrans
, ac
->root
);
1485 int btrfs_commit_transaction_async(struct btrfs_trans_handle
*trans
,
1486 struct btrfs_root
*root
,
1487 int wait_for_unblock
)
1489 struct btrfs_async_commit
*ac
;
1490 struct btrfs_transaction
*cur_trans
;
1492 ac
= kmalloc(sizeof(*ac
), GFP_NOFS
);
1496 INIT_WORK(&ac
->work
, do_async_commit
);
1498 ac
->newtrans
= btrfs_join_transaction(root
);
1499 if (IS_ERR(ac
->newtrans
)) {
1500 int err
= PTR_ERR(ac
->newtrans
);
1505 /* take transaction reference */
1506 cur_trans
= trans
->transaction
;
1507 atomic_inc(&cur_trans
->use_count
);
1509 btrfs_end_transaction(trans
, root
);
1512 * Tell lockdep we've released the freeze rwsem, since the
1513 * async commit thread will be the one to unlock it.
1515 if (ac
->newtrans
->type
& __TRANS_FREEZABLE
)
1517 &root
->fs_info
->sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
1520 schedule_work(&ac
->work
);
1522 /* wait for transaction to start and unblock */
1523 if (wait_for_unblock
)
1524 wait_current_trans_commit_start_and_unblock(root
, cur_trans
);
1526 wait_current_trans_commit_start(root
, cur_trans
);
1528 if (current
->journal_info
== trans
)
1529 current
->journal_info
= NULL
;
1531 btrfs_put_transaction(cur_trans
);
1536 static void cleanup_transaction(struct btrfs_trans_handle
*trans
,
1537 struct btrfs_root
*root
, int err
)
1539 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1542 WARN_ON(trans
->use_count
> 1);
1544 btrfs_abort_transaction(trans
, root
, err
);
1546 spin_lock(&root
->fs_info
->trans_lock
);
1549 * If the transaction is removed from the list, it means this
1550 * transaction has been committed successfully, so it is impossible
1551 * to call the cleanup function.
1553 BUG_ON(list_empty(&cur_trans
->list
));
1555 list_del_init(&cur_trans
->list
);
1556 if (cur_trans
== root
->fs_info
->running_transaction
) {
1557 cur_trans
->state
= TRANS_STATE_COMMIT_DOING
;
1558 spin_unlock(&root
->fs_info
->trans_lock
);
1559 wait_event(cur_trans
->writer_wait
,
1560 atomic_read(&cur_trans
->num_writers
) == 1);
1562 spin_lock(&root
->fs_info
->trans_lock
);
1564 spin_unlock(&root
->fs_info
->trans_lock
);
1566 btrfs_cleanup_one_transaction(trans
->transaction
, root
);
1568 spin_lock(&root
->fs_info
->trans_lock
);
1569 if (cur_trans
== root
->fs_info
->running_transaction
)
1570 root
->fs_info
->running_transaction
= NULL
;
1571 spin_unlock(&root
->fs_info
->trans_lock
);
1573 if (trans
->type
& __TRANS_FREEZABLE
)
1574 sb_end_intwrite(root
->fs_info
->sb
);
1575 btrfs_put_transaction(cur_trans
);
1576 btrfs_put_transaction(cur_trans
);
1578 trace_btrfs_transaction_commit(root
);
1580 btrfs_scrub_continue(root
);
1582 if (current
->journal_info
== trans
)
1583 current
->journal_info
= NULL
;
1585 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1588 static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle
*trans
,
1589 struct btrfs_root
*root
)
1593 ret
= btrfs_run_delayed_items(trans
, root
);
1598 * running the delayed items may have added new refs. account
1599 * them now so that they hinder processing of more delayed refs
1600 * as little as possible.
1602 btrfs_delayed_refs_qgroup_accounting(trans
, root
->fs_info
);
1605 * rename don't use btrfs_join_transaction, so, once we
1606 * set the transaction to blocked above, we aren't going
1607 * to get any new ordered operations. We can safely run
1608 * it here and no for sure that nothing new will be added
1611 ret
= btrfs_run_ordered_operations(trans
, root
, 1);
1616 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info
*fs_info
)
1618 if (btrfs_test_opt(fs_info
->tree_root
, FLUSHONCOMMIT
))
1619 return btrfs_start_all_delalloc_inodes(fs_info
, 1);
1623 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info
*fs_info
)
1625 if (btrfs_test_opt(fs_info
->tree_root
, FLUSHONCOMMIT
))
1626 btrfs_wait_all_ordered_extents(fs_info
);
1629 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
,
1630 struct btrfs_root
*root
)
1632 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1633 struct btrfs_transaction
*prev_trans
= NULL
;
1636 ret
= btrfs_run_ordered_operations(trans
, root
, 0);
1638 btrfs_abort_transaction(trans
, root
, ret
);
1639 btrfs_end_transaction(trans
, root
);
1643 /* Stop the commit early if ->aborted is set */
1644 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1645 ret
= cur_trans
->aborted
;
1646 btrfs_end_transaction(trans
, root
);
1650 /* make a pass through all the delayed refs we have so far
1651 * any runnings procs may add more while we are here
1653 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1655 btrfs_end_transaction(trans
, root
);
1659 btrfs_trans_release_metadata(trans
, root
);
1660 trans
->block_rsv
= NULL
;
1661 if (trans
->qgroup_reserved
) {
1662 btrfs_qgroup_free(root
, trans
->qgroup_reserved
);
1663 trans
->qgroup_reserved
= 0;
1666 cur_trans
= trans
->transaction
;
1669 * set the flushing flag so procs in this transaction have to
1670 * start sending their work down.
1672 cur_trans
->delayed_refs
.flushing
= 1;
1675 if (!list_empty(&trans
->new_bgs
))
1676 btrfs_create_pending_block_groups(trans
, root
);
1678 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1680 btrfs_end_transaction(trans
, root
);
1684 spin_lock(&root
->fs_info
->trans_lock
);
1685 if (cur_trans
->state
>= TRANS_STATE_COMMIT_START
) {
1686 spin_unlock(&root
->fs_info
->trans_lock
);
1687 atomic_inc(&cur_trans
->use_count
);
1688 ret
= btrfs_end_transaction(trans
, root
);
1690 wait_for_commit(root
, cur_trans
);
1692 btrfs_put_transaction(cur_trans
);
1697 cur_trans
->state
= TRANS_STATE_COMMIT_START
;
1698 wake_up(&root
->fs_info
->transaction_blocked_wait
);
1700 if (cur_trans
->list
.prev
!= &root
->fs_info
->trans_list
) {
1701 prev_trans
= list_entry(cur_trans
->list
.prev
,
1702 struct btrfs_transaction
, list
);
1703 if (prev_trans
->state
!= TRANS_STATE_COMPLETED
) {
1704 atomic_inc(&prev_trans
->use_count
);
1705 spin_unlock(&root
->fs_info
->trans_lock
);
1707 wait_for_commit(root
, prev_trans
);
1709 btrfs_put_transaction(prev_trans
);
1711 spin_unlock(&root
->fs_info
->trans_lock
);
1714 spin_unlock(&root
->fs_info
->trans_lock
);
1717 extwriter_counter_dec(cur_trans
, trans
->type
);
1719 ret
= btrfs_start_delalloc_flush(root
->fs_info
);
1721 goto cleanup_transaction
;
1723 ret
= btrfs_flush_all_pending_stuffs(trans
, root
);
1725 goto cleanup_transaction
;
1727 wait_event(cur_trans
->writer_wait
,
1728 extwriter_counter_read(cur_trans
) == 0);
1730 /* some pending stuffs might be added after the previous flush. */
1731 ret
= btrfs_flush_all_pending_stuffs(trans
, root
);
1733 goto cleanup_transaction
;
1735 btrfs_wait_delalloc_flush(root
->fs_info
);
1737 * Ok now we need to make sure to block out any other joins while we
1738 * commit the transaction. We could have started a join before setting
1739 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
1741 spin_lock(&root
->fs_info
->trans_lock
);
1742 cur_trans
->state
= TRANS_STATE_COMMIT_DOING
;
1743 spin_unlock(&root
->fs_info
->trans_lock
);
1744 wait_event(cur_trans
->writer_wait
,
1745 atomic_read(&cur_trans
->num_writers
) == 1);
1747 /* ->aborted might be set after the previous check, so check it */
1748 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1749 ret
= cur_trans
->aborted
;
1750 goto cleanup_transaction
;
1753 * the reloc mutex makes sure that we stop
1754 * the balancing code from coming in and moving
1755 * extents around in the middle of the commit
1757 mutex_lock(&root
->fs_info
->reloc_mutex
);
1760 * We needn't worry about the delayed items because we will
1761 * deal with them in create_pending_snapshot(), which is the
1762 * core function of the snapshot creation.
1764 ret
= create_pending_snapshots(trans
, root
->fs_info
);
1766 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1767 goto cleanup_transaction
;
1771 * We insert the dir indexes of the snapshots and update the inode
1772 * of the snapshots' parents after the snapshot creation, so there
1773 * are some delayed items which are not dealt with. Now deal with
1776 * We needn't worry that this operation will corrupt the snapshots,
1777 * because all the tree which are snapshoted will be forced to COW
1778 * the nodes and leaves.
1780 ret
= btrfs_run_delayed_items(trans
, root
);
1782 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1783 goto cleanup_transaction
;
1786 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1788 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1789 goto cleanup_transaction
;
1793 * make sure none of the code above managed to slip in a
1796 btrfs_assert_delayed_root_empty(root
);
1798 WARN_ON(cur_trans
!= trans
->transaction
);
1800 btrfs_scrub_pause(root
);
1801 /* btrfs_commit_tree_roots is responsible for getting the
1802 * various roots consistent with each other. Every pointer
1803 * in the tree of tree roots has to point to the most up to date
1804 * root for every subvolume and other tree. So, we have to keep
1805 * the tree logging code from jumping in and changing any
1808 * At this point in the commit, there can't be any tree-log
1809 * writers, but a little lower down we drop the trans mutex
1810 * and let new people in. By holding the tree_log_mutex
1811 * from now until after the super is written, we avoid races
1812 * with the tree-log code.
1814 mutex_lock(&root
->fs_info
->tree_log_mutex
);
1816 ret
= commit_fs_roots(trans
, root
);
1818 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1819 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1820 goto cleanup_transaction
;
1823 /* commit_fs_roots gets rid of all the tree log roots, it is now
1824 * safe to free the root of tree log roots
1826 btrfs_free_log_root_tree(trans
, root
->fs_info
);
1828 ret
= commit_cowonly_roots(trans
, root
);
1830 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1831 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1832 goto cleanup_transaction
;
1836 * The tasks which save the space cache and inode cache may also
1837 * update ->aborted, check it.
1839 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1840 ret
= cur_trans
->aborted
;
1841 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1842 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1843 goto cleanup_transaction
;
1846 btrfs_prepare_extent_commit(trans
, root
);
1848 cur_trans
= root
->fs_info
->running_transaction
;
1850 btrfs_set_root_node(&root
->fs_info
->tree_root
->root_item
,
1851 root
->fs_info
->tree_root
->node
);
1852 switch_commit_root(root
->fs_info
->tree_root
);
1854 btrfs_set_root_node(&root
->fs_info
->chunk_root
->root_item
,
1855 root
->fs_info
->chunk_root
->node
);
1856 switch_commit_root(root
->fs_info
->chunk_root
);
1858 assert_qgroups_uptodate(trans
);
1859 update_super_roots(root
);
1861 btrfs_set_super_log_root(root
->fs_info
->super_copy
, 0);
1862 btrfs_set_super_log_root_level(root
->fs_info
->super_copy
, 0);
1863 memcpy(root
->fs_info
->super_for_commit
, root
->fs_info
->super_copy
,
1864 sizeof(*root
->fs_info
->super_copy
));
1866 spin_lock(&root
->fs_info
->trans_lock
);
1867 cur_trans
->state
= TRANS_STATE_UNBLOCKED
;
1868 root
->fs_info
->running_transaction
= NULL
;
1869 spin_unlock(&root
->fs_info
->trans_lock
);
1870 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1872 wake_up(&root
->fs_info
->transaction_wait
);
1874 ret
= btrfs_write_and_wait_transaction(trans
, root
);
1876 btrfs_error(root
->fs_info
, ret
,
1877 "Error while writing out transaction");
1878 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1879 goto cleanup_transaction
;
1882 ret
= write_ctree_super(trans
, root
, 0);
1884 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1885 goto cleanup_transaction
;
1889 * the super is written, we can safely allow the tree-loggers
1890 * to go about their business
1892 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1894 btrfs_finish_extent_commit(trans
, root
);
1896 root
->fs_info
->last_trans_committed
= cur_trans
->transid
;
1898 * We needn't acquire the lock here because there is no other task
1899 * which can change it.
1901 cur_trans
->state
= TRANS_STATE_COMPLETED
;
1902 wake_up(&cur_trans
->commit_wait
);
1904 spin_lock(&root
->fs_info
->trans_lock
);
1905 list_del_init(&cur_trans
->list
);
1906 spin_unlock(&root
->fs_info
->trans_lock
);
1908 btrfs_put_transaction(cur_trans
);
1909 btrfs_put_transaction(cur_trans
);
1911 if (trans
->type
& __TRANS_FREEZABLE
)
1912 sb_end_intwrite(root
->fs_info
->sb
);
1914 trace_btrfs_transaction_commit(root
);
1916 btrfs_scrub_continue(root
);
1918 if (current
->journal_info
== trans
)
1919 current
->journal_info
= NULL
;
1921 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1923 if (current
!= root
->fs_info
->transaction_kthread
)
1924 btrfs_run_delayed_iputs(root
);
1928 cleanup_transaction
:
1929 btrfs_trans_release_metadata(trans
, root
);
1930 trans
->block_rsv
= NULL
;
1931 if (trans
->qgroup_reserved
) {
1932 btrfs_qgroup_free(root
, trans
->qgroup_reserved
);
1933 trans
->qgroup_reserved
= 0;
1935 btrfs_warn(root
->fs_info
, "Skipping commit of aborted transaction.");
1936 if (current
->journal_info
== trans
)
1937 current
->journal_info
= NULL
;
1938 cleanup_transaction(trans
, root
, ret
);
1944 * return < 0 if error
1945 * 0 if there are no more dead_roots at the time of call
1946 * 1 there are more to be processed, call me again
1948 * The return value indicates there are certainly more snapshots to delete, but
1949 * if there comes a new one during processing, it may return 0. We don't mind,
1950 * because btrfs_commit_super will poke cleaner thread and it will process it a
1951 * few seconds later.
1953 int btrfs_clean_one_deleted_snapshot(struct btrfs_root
*root
)
1956 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1958 spin_lock(&fs_info
->trans_lock
);
1959 if (list_empty(&fs_info
->dead_roots
)) {
1960 spin_unlock(&fs_info
->trans_lock
);
1963 root
= list_first_entry(&fs_info
->dead_roots
,
1964 struct btrfs_root
, root_list
);
1965 list_del_init(&root
->root_list
);
1966 spin_unlock(&fs_info
->trans_lock
);
1968 pr_debug("btrfs: cleaner removing %llu\n", root
->objectid
);
1970 btrfs_kill_all_delayed_nodes(root
);
1972 if (btrfs_header_backref_rev(root
->node
) <
1973 BTRFS_MIXED_BACKREF_REV
)
1974 ret
= btrfs_drop_snapshot(root
, NULL
, 0, 0);
1976 ret
= btrfs_drop_snapshot(root
, NULL
, 1, 0);
1978 * If we encounter a transaction abort during snapshot cleaning, we
1979 * don't want to crash here
1981 return (ret
< 0) ? 0 : 1;