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"
36 #define BTRFS_ROOT_TRANS_TAG 0
38 static unsigned int btrfs_blocked_trans_types
[TRANS_STATE_MAX
] = {
39 [TRANS_STATE_RUNNING
] = 0U,
40 [TRANS_STATE_BLOCKED
] = (__TRANS_USERSPACE
|
42 [TRANS_STATE_COMMIT_START
] = (__TRANS_USERSPACE
|
45 [TRANS_STATE_COMMIT_DOING
] = (__TRANS_USERSPACE
|
49 [TRANS_STATE_UNBLOCKED
] = (__TRANS_USERSPACE
|
54 [TRANS_STATE_COMPLETED
] = (__TRANS_USERSPACE
|
61 void btrfs_put_transaction(struct btrfs_transaction
*transaction
)
63 WARN_ON(atomic_read(&transaction
->use_count
) == 0);
64 if (atomic_dec_and_test(&transaction
->use_count
)) {
65 BUG_ON(!list_empty(&transaction
->list
));
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_roots(struct btrfs_transaction
*trans
,
80 struct btrfs_fs_info
*fs_info
)
82 struct btrfs_root
*root
, *tmp
;
84 down_write(&fs_info
->commit_root_sem
);
85 list_for_each_entry_safe(root
, tmp
, &trans
->switch_commits
,
87 list_del_init(&root
->dirty_list
);
88 free_extent_buffer(root
->commit_root
);
89 root
->commit_root
= btrfs_root_node(root
);
90 if (is_fstree(root
->objectid
))
91 btrfs_unpin_free_ino(root
);
93 up_write(&fs_info
->commit_root_sem
);
96 static inline void extwriter_counter_inc(struct btrfs_transaction
*trans
,
99 if (type
& TRANS_EXTWRITERS
)
100 atomic_inc(&trans
->num_extwriters
);
103 static inline void extwriter_counter_dec(struct btrfs_transaction
*trans
,
106 if (type
& TRANS_EXTWRITERS
)
107 atomic_dec(&trans
->num_extwriters
);
110 static inline void extwriter_counter_init(struct btrfs_transaction
*trans
,
113 atomic_set(&trans
->num_extwriters
, ((type
& TRANS_EXTWRITERS
) ? 1 : 0));
116 static inline int extwriter_counter_read(struct btrfs_transaction
*trans
)
118 return atomic_read(&trans
->num_extwriters
);
122 * either allocate a new transaction or hop into the existing one
124 static noinline
int join_transaction(struct btrfs_root
*root
, unsigned int type
)
126 struct btrfs_transaction
*cur_trans
;
127 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
129 spin_lock(&fs_info
->trans_lock
);
131 /* The file system has been taken offline. No new transactions. */
132 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
133 spin_unlock(&fs_info
->trans_lock
);
137 cur_trans
= fs_info
->running_transaction
;
139 if (cur_trans
->aborted
) {
140 spin_unlock(&fs_info
->trans_lock
);
141 return cur_trans
->aborted
;
143 if (btrfs_blocked_trans_types
[cur_trans
->state
] & type
) {
144 spin_unlock(&fs_info
->trans_lock
);
147 atomic_inc(&cur_trans
->use_count
);
148 atomic_inc(&cur_trans
->num_writers
);
149 extwriter_counter_inc(cur_trans
, type
);
150 spin_unlock(&fs_info
->trans_lock
);
153 spin_unlock(&fs_info
->trans_lock
);
156 * If we are ATTACH, we just want to catch the current transaction,
157 * and commit it. If there is no transaction, just return ENOENT.
159 if (type
== TRANS_ATTACH
)
163 * JOIN_NOLOCK only happens during the transaction commit, so
164 * it is impossible that ->running_transaction is NULL
166 BUG_ON(type
== TRANS_JOIN_NOLOCK
);
168 cur_trans
= kmem_cache_alloc(btrfs_transaction_cachep
, GFP_NOFS
);
172 spin_lock(&fs_info
->trans_lock
);
173 if (fs_info
->running_transaction
) {
175 * someone started a transaction after we unlocked. Make sure
176 * to redo the checks above
178 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
180 } else if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
181 spin_unlock(&fs_info
->trans_lock
);
182 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
186 atomic_set(&cur_trans
->num_writers
, 1);
187 extwriter_counter_init(cur_trans
, type
);
188 init_waitqueue_head(&cur_trans
->writer_wait
);
189 init_waitqueue_head(&cur_trans
->commit_wait
);
190 cur_trans
->state
= TRANS_STATE_RUNNING
;
192 * One for this trans handle, one so it will live on until we
193 * commit the transaction.
195 atomic_set(&cur_trans
->use_count
, 2);
196 cur_trans
->start_time
= get_seconds();
198 cur_trans
->delayed_refs
.href_root
= RB_ROOT
;
199 atomic_set(&cur_trans
->delayed_refs
.num_entries
, 0);
200 cur_trans
->delayed_refs
.num_heads_ready
= 0;
201 cur_trans
->delayed_refs
.num_heads
= 0;
202 cur_trans
->delayed_refs
.flushing
= 0;
203 cur_trans
->delayed_refs
.run_delayed_start
= 0;
206 * although the tree mod log is per file system and not per transaction,
207 * the log must never go across transaction boundaries.
210 if (!list_empty(&fs_info
->tree_mod_seq_list
))
211 WARN(1, KERN_ERR
"BTRFS: tree_mod_seq_list not empty when "
212 "creating a fresh transaction\n");
213 if (!RB_EMPTY_ROOT(&fs_info
->tree_mod_log
))
214 WARN(1, KERN_ERR
"BTRFS: tree_mod_log rb tree not empty when "
215 "creating a fresh transaction\n");
216 atomic64_set(&fs_info
->tree_mod_seq
, 0);
218 spin_lock_init(&cur_trans
->delayed_refs
.lock
);
220 INIT_LIST_HEAD(&cur_trans
->pending_snapshots
);
221 INIT_LIST_HEAD(&cur_trans
->ordered_operations
);
222 INIT_LIST_HEAD(&cur_trans
->pending_chunks
);
223 INIT_LIST_HEAD(&cur_trans
->switch_commits
);
224 list_add_tail(&cur_trans
->list
, &fs_info
->trans_list
);
225 extent_io_tree_init(&cur_trans
->dirty_pages
,
226 fs_info
->btree_inode
->i_mapping
);
227 fs_info
->generation
++;
228 cur_trans
->transid
= fs_info
->generation
;
229 fs_info
->running_transaction
= cur_trans
;
230 cur_trans
->aborted
= 0;
231 spin_unlock(&fs_info
->trans_lock
);
237 * this does all the record keeping required to make sure that a reference
238 * counted root is properly recorded in a given transaction. This is required
239 * to make sure the old root from before we joined the transaction is deleted
240 * when the transaction commits
242 static int record_root_in_trans(struct btrfs_trans_handle
*trans
,
243 struct btrfs_root
*root
)
245 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) &&
246 root
->last_trans
< trans
->transid
) {
247 WARN_ON(root
== root
->fs_info
->extent_root
);
248 WARN_ON(root
->commit_root
!= root
->node
);
251 * see below for IN_TRANS_SETUP usage rules
252 * we have the reloc mutex held now, so there
253 * is only one writer in this function
255 set_bit(BTRFS_ROOT_IN_TRANS_SETUP
, &root
->state
);
257 /* make sure readers find IN_TRANS_SETUP before
258 * they find our root->last_trans update
262 spin_lock(&root
->fs_info
->fs_roots_radix_lock
);
263 if (root
->last_trans
== trans
->transid
) {
264 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
267 radix_tree_tag_set(&root
->fs_info
->fs_roots_radix
,
268 (unsigned long)root
->root_key
.objectid
,
269 BTRFS_ROOT_TRANS_TAG
);
270 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
271 root
->last_trans
= trans
->transid
;
273 /* this is pretty tricky. We don't want to
274 * take the relocation lock in btrfs_record_root_in_trans
275 * unless we're really doing the first setup for this root in
278 * Normally we'd use root->last_trans as a flag to decide
279 * if we want to take the expensive mutex.
281 * But, we have to set root->last_trans before we
282 * init the relocation root, otherwise, we trip over warnings
283 * in ctree.c. The solution used here is to flag ourselves
284 * with root IN_TRANS_SETUP. When this is 1, we're still
285 * fixing up the reloc trees and everyone must wait.
287 * When this is zero, they can trust root->last_trans and fly
288 * through btrfs_record_root_in_trans without having to take the
289 * lock. smp_wmb() makes sure that all the writes above are
290 * done before we pop in the zero below
292 btrfs_init_reloc_root(trans
, root
);
293 smp_mb__before_atomic();
294 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP
, &root
->state
);
300 int btrfs_record_root_in_trans(struct btrfs_trans_handle
*trans
,
301 struct btrfs_root
*root
)
303 if (!test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
307 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
311 if (root
->last_trans
== trans
->transid
&&
312 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP
, &root
->state
))
315 mutex_lock(&root
->fs_info
->reloc_mutex
);
316 record_root_in_trans(trans
, root
);
317 mutex_unlock(&root
->fs_info
->reloc_mutex
);
322 static inline int is_transaction_blocked(struct btrfs_transaction
*trans
)
324 return (trans
->state
>= TRANS_STATE_BLOCKED
&&
325 trans
->state
< TRANS_STATE_UNBLOCKED
&&
329 /* wait for commit against the current transaction to become unblocked
330 * when this is done, it is safe to start a new transaction, but the current
331 * transaction might not be fully on disk.
333 static void wait_current_trans(struct btrfs_root
*root
)
335 struct btrfs_transaction
*cur_trans
;
337 spin_lock(&root
->fs_info
->trans_lock
);
338 cur_trans
= root
->fs_info
->running_transaction
;
339 if (cur_trans
&& is_transaction_blocked(cur_trans
)) {
340 atomic_inc(&cur_trans
->use_count
);
341 spin_unlock(&root
->fs_info
->trans_lock
);
343 wait_event(root
->fs_info
->transaction_wait
,
344 cur_trans
->state
>= TRANS_STATE_UNBLOCKED
||
346 btrfs_put_transaction(cur_trans
);
348 spin_unlock(&root
->fs_info
->trans_lock
);
352 static int may_wait_transaction(struct btrfs_root
*root
, int type
)
354 if (root
->fs_info
->log_root_recovering
)
357 if (type
== TRANS_USERSPACE
)
360 if (type
== TRANS_START
&&
361 !atomic_read(&root
->fs_info
->open_ioctl_trans
))
367 static inline bool need_reserve_reloc_root(struct btrfs_root
*root
)
369 if (!root
->fs_info
->reloc_ctl
||
370 !test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
371 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
378 static struct btrfs_trans_handle
*
379 start_transaction(struct btrfs_root
*root
, u64 num_items
, unsigned int type
,
380 enum btrfs_reserve_flush_enum flush
)
382 struct btrfs_trans_handle
*h
;
383 struct btrfs_transaction
*cur_trans
;
385 u64 qgroup_reserved
= 0;
386 bool reloc_reserved
= false;
389 /* Send isn't supposed to start transactions. */
390 ASSERT(current
->journal_info
!= (void *)BTRFS_SEND_TRANS_STUB
);
392 if (test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
393 return ERR_PTR(-EROFS
);
395 if (current
->journal_info
) {
396 WARN_ON(type
& TRANS_EXTWRITERS
);
397 h
= current
->journal_info
;
399 WARN_ON(h
->use_count
> 2);
400 h
->orig_rsv
= h
->block_rsv
;
406 * Do the reservation before we join the transaction so we can do all
407 * the appropriate flushing if need be.
409 if (num_items
> 0 && root
!= root
->fs_info
->chunk_root
) {
410 if (root
->fs_info
->quota_enabled
&&
411 is_fstree(root
->root_key
.objectid
)) {
412 qgroup_reserved
= num_items
* root
->leafsize
;
413 ret
= btrfs_qgroup_reserve(root
, qgroup_reserved
);
418 num_bytes
= btrfs_calc_trans_metadata_size(root
, num_items
);
420 * Do the reservation for the relocation root creation
422 if (unlikely(need_reserve_reloc_root(root
))) {
423 num_bytes
+= root
->nodesize
;
424 reloc_reserved
= true;
427 ret
= btrfs_block_rsv_add(root
,
428 &root
->fs_info
->trans_block_rsv
,
434 h
= kmem_cache_alloc(btrfs_trans_handle_cachep
, GFP_NOFS
);
441 * If we are JOIN_NOLOCK we're already committing a transaction and
442 * waiting on this guy, so we don't need to do the sb_start_intwrite
443 * because we're already holding a ref. We need this because we could
444 * have raced in and did an fsync() on a file which can kick a commit
445 * and then we deadlock with somebody doing a freeze.
447 * If we are ATTACH, it means we just want to catch the current
448 * transaction and commit it, so we needn't do sb_start_intwrite().
450 if (type
& __TRANS_FREEZABLE
)
451 sb_start_intwrite(root
->fs_info
->sb
);
453 if (may_wait_transaction(root
, type
))
454 wait_current_trans(root
);
457 ret
= join_transaction(root
, type
);
459 wait_current_trans(root
);
460 if (unlikely(type
== TRANS_ATTACH
))
463 } while (ret
== -EBUSY
);
466 /* We must get the transaction if we are JOIN_NOLOCK. */
467 BUG_ON(type
== TRANS_JOIN_NOLOCK
);
471 cur_trans
= root
->fs_info
->running_transaction
;
473 h
->transid
= cur_trans
->transid
;
474 h
->transaction
= cur_trans
;
476 h
->bytes_reserved
= 0;
478 h
->delayed_ref_updates
= 0;
484 h
->qgroup_reserved
= 0;
485 h
->delayed_ref_elem
.seq
= 0;
487 h
->allocating_chunk
= false;
488 h
->reloc_reserved
= false;
490 INIT_LIST_HEAD(&h
->qgroup_ref_list
);
491 INIT_LIST_HEAD(&h
->new_bgs
);
494 if (cur_trans
->state
>= TRANS_STATE_BLOCKED
&&
495 may_wait_transaction(root
, type
)) {
496 current
->journal_info
= h
;
497 btrfs_commit_transaction(h
, root
);
502 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
503 h
->transid
, num_bytes
, 1);
504 h
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
505 h
->bytes_reserved
= num_bytes
;
506 h
->reloc_reserved
= reloc_reserved
;
508 h
->qgroup_reserved
= qgroup_reserved
;
511 btrfs_record_root_in_trans(h
, root
);
513 if (!current
->journal_info
&& type
!= TRANS_USERSPACE
)
514 current
->journal_info
= h
;
518 if (type
& __TRANS_FREEZABLE
)
519 sb_end_intwrite(root
->fs_info
->sb
);
520 kmem_cache_free(btrfs_trans_handle_cachep
, h
);
523 btrfs_block_rsv_release(root
, &root
->fs_info
->trans_block_rsv
,
527 btrfs_qgroup_free(root
, qgroup_reserved
);
531 struct btrfs_trans_handle
*btrfs_start_transaction(struct btrfs_root
*root
,
534 return start_transaction(root
, num_items
, TRANS_START
,
535 BTRFS_RESERVE_FLUSH_ALL
);
538 struct btrfs_trans_handle
*btrfs_start_transaction_lflush(
539 struct btrfs_root
*root
, int num_items
)
541 return start_transaction(root
, num_items
, TRANS_START
,
542 BTRFS_RESERVE_FLUSH_LIMIT
);
545 struct btrfs_trans_handle
*btrfs_join_transaction(struct btrfs_root
*root
)
547 return start_transaction(root
, 0, TRANS_JOIN
, 0);
550 struct btrfs_trans_handle
*btrfs_join_transaction_nolock(struct btrfs_root
*root
)
552 return start_transaction(root
, 0, TRANS_JOIN_NOLOCK
, 0);
555 struct btrfs_trans_handle
*btrfs_start_ioctl_transaction(struct btrfs_root
*root
)
557 return start_transaction(root
, 0, TRANS_USERSPACE
, 0);
561 * btrfs_attach_transaction() - catch the running transaction
563 * It is used when we want to commit the current the transaction, but
564 * don't want to start a new one.
566 * Note: If this function return -ENOENT, it just means there is no
567 * running transaction. But it is possible that the inactive transaction
568 * is still in the memory, not fully on disk. If you hope there is no
569 * inactive transaction in the fs when -ENOENT is returned, you should
571 * btrfs_attach_transaction_barrier()
573 struct btrfs_trans_handle
*btrfs_attach_transaction(struct btrfs_root
*root
)
575 return start_transaction(root
, 0, TRANS_ATTACH
, 0);
579 * btrfs_attach_transaction_barrier() - catch the running transaction
581 * It is similar to the above function, the differentia is this one
582 * will wait for all the inactive transactions until they fully
585 struct btrfs_trans_handle
*
586 btrfs_attach_transaction_barrier(struct btrfs_root
*root
)
588 struct btrfs_trans_handle
*trans
;
590 trans
= start_transaction(root
, 0, TRANS_ATTACH
, 0);
591 if (IS_ERR(trans
) && PTR_ERR(trans
) == -ENOENT
)
592 btrfs_wait_for_commit(root
, 0);
597 /* wait for a transaction commit to be fully complete */
598 static noinline
void wait_for_commit(struct btrfs_root
*root
,
599 struct btrfs_transaction
*commit
)
601 wait_event(commit
->commit_wait
, commit
->state
== TRANS_STATE_COMPLETED
);
604 int btrfs_wait_for_commit(struct btrfs_root
*root
, u64 transid
)
606 struct btrfs_transaction
*cur_trans
= NULL
, *t
;
610 if (transid
<= root
->fs_info
->last_trans_committed
)
614 /* find specified transaction */
615 spin_lock(&root
->fs_info
->trans_lock
);
616 list_for_each_entry(t
, &root
->fs_info
->trans_list
, list
) {
617 if (t
->transid
== transid
) {
619 atomic_inc(&cur_trans
->use_count
);
623 if (t
->transid
> transid
) {
628 spin_unlock(&root
->fs_info
->trans_lock
);
629 /* The specified transaction doesn't exist */
633 /* find newest transaction that is committing | committed */
634 spin_lock(&root
->fs_info
->trans_lock
);
635 list_for_each_entry_reverse(t
, &root
->fs_info
->trans_list
,
637 if (t
->state
>= TRANS_STATE_COMMIT_START
) {
638 if (t
->state
== TRANS_STATE_COMPLETED
)
641 atomic_inc(&cur_trans
->use_count
);
645 spin_unlock(&root
->fs_info
->trans_lock
);
647 goto out
; /* nothing committing|committed */
650 wait_for_commit(root
, cur_trans
);
651 btrfs_put_transaction(cur_trans
);
656 void btrfs_throttle(struct btrfs_root
*root
)
658 if (!atomic_read(&root
->fs_info
->open_ioctl_trans
))
659 wait_current_trans(root
);
662 static int should_end_transaction(struct btrfs_trans_handle
*trans
,
663 struct btrfs_root
*root
)
665 if (root
->fs_info
->global_block_rsv
.space_info
->full
&&
666 btrfs_check_space_for_delayed_refs(trans
, root
))
669 return !!btrfs_block_rsv_check(root
, &root
->fs_info
->global_block_rsv
, 5);
672 int btrfs_should_end_transaction(struct btrfs_trans_handle
*trans
,
673 struct btrfs_root
*root
)
675 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
680 if (cur_trans
->state
>= TRANS_STATE_BLOCKED
||
681 cur_trans
->delayed_refs
.flushing
)
684 updates
= trans
->delayed_ref_updates
;
685 trans
->delayed_ref_updates
= 0;
687 err
= btrfs_run_delayed_refs(trans
, root
, updates
);
688 if (err
) /* Error code will also eval true */
692 return should_end_transaction(trans
, root
);
695 static int __btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
696 struct btrfs_root
*root
, int throttle
)
698 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
699 struct btrfs_fs_info
*info
= root
->fs_info
;
700 unsigned long cur
= trans
->delayed_ref_updates
;
701 int lock
= (trans
->type
!= TRANS_JOIN_NOLOCK
);
703 int must_run_delayed_refs
= 0;
705 if (trans
->use_count
> 1) {
707 trans
->block_rsv
= trans
->orig_rsv
;
711 btrfs_trans_release_metadata(trans
, root
);
712 trans
->block_rsv
= NULL
;
714 if (!list_empty(&trans
->new_bgs
))
715 btrfs_create_pending_block_groups(trans
, root
);
717 trans
->delayed_ref_updates
= 0;
719 must_run_delayed_refs
=
720 btrfs_should_throttle_delayed_refs(trans
, root
);
721 cur
= max_t(unsigned long, cur
, 32);
724 * don't make the caller wait if they are from a NOLOCK
725 * or ATTACH transaction, it will deadlock with commit
727 if (must_run_delayed_refs
== 1 &&
728 (trans
->type
& (__TRANS_JOIN_NOLOCK
| __TRANS_ATTACH
)))
729 must_run_delayed_refs
= 2;
732 if (trans
->qgroup_reserved
) {
734 * the same root has to be passed here between start_transaction
735 * and end_transaction. Subvolume quota depends on this.
737 btrfs_qgroup_free(trans
->root
, trans
->qgroup_reserved
);
738 trans
->qgroup_reserved
= 0;
741 btrfs_trans_release_metadata(trans
, root
);
742 trans
->block_rsv
= NULL
;
744 if (!list_empty(&trans
->new_bgs
))
745 btrfs_create_pending_block_groups(trans
, root
);
747 if (lock
&& !atomic_read(&root
->fs_info
->open_ioctl_trans
) &&
748 should_end_transaction(trans
, root
) &&
749 ACCESS_ONCE(cur_trans
->state
) == TRANS_STATE_RUNNING
) {
750 spin_lock(&info
->trans_lock
);
751 if (cur_trans
->state
== TRANS_STATE_RUNNING
)
752 cur_trans
->state
= TRANS_STATE_BLOCKED
;
753 spin_unlock(&info
->trans_lock
);
756 if (lock
&& ACCESS_ONCE(cur_trans
->state
) == TRANS_STATE_BLOCKED
) {
758 return btrfs_commit_transaction(trans
, root
);
760 wake_up_process(info
->transaction_kthread
);
763 if (trans
->type
& __TRANS_FREEZABLE
)
764 sb_end_intwrite(root
->fs_info
->sb
);
766 WARN_ON(cur_trans
!= info
->running_transaction
);
767 WARN_ON(atomic_read(&cur_trans
->num_writers
) < 1);
768 atomic_dec(&cur_trans
->num_writers
);
769 extwriter_counter_dec(cur_trans
, trans
->type
);
772 if (waitqueue_active(&cur_trans
->writer_wait
))
773 wake_up(&cur_trans
->writer_wait
);
774 btrfs_put_transaction(cur_trans
);
776 if (current
->journal_info
== trans
)
777 current
->journal_info
= NULL
;
780 btrfs_run_delayed_iputs(root
);
782 if (trans
->aborted
||
783 test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
)) {
784 wake_up_process(info
->transaction_kthread
);
787 assert_qgroups_uptodate(trans
);
789 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
790 if (must_run_delayed_refs
) {
791 btrfs_async_run_delayed_refs(root
, cur
,
792 must_run_delayed_refs
== 1);
797 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
798 struct btrfs_root
*root
)
800 return __btrfs_end_transaction(trans
, root
, 0);
803 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
,
804 struct btrfs_root
*root
)
806 return __btrfs_end_transaction(trans
, root
, 1);
810 * when btree blocks are allocated, they have some corresponding bits set for
811 * them in one of two extent_io trees. This is used to make sure all of
812 * those extents are sent to disk but does not wait on them
814 int btrfs_write_marked_extents(struct btrfs_root
*root
,
815 struct extent_io_tree
*dirty_pages
, int mark
)
819 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
820 struct extent_state
*cached_state
= NULL
;
824 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
825 mark
, &cached_state
)) {
826 convert_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
,
827 mark
, &cached_state
, GFP_NOFS
);
829 err
= filemap_fdatawrite_range(mapping
, start
, end
);
841 * when btree blocks are allocated, they have some corresponding bits set for
842 * them in one of two extent_io trees. This is used to make sure all of
843 * those extents are on disk for transaction or log commit. We wait
844 * on all the pages and clear them from the dirty pages state tree
846 int btrfs_wait_marked_extents(struct btrfs_root
*root
,
847 struct extent_io_tree
*dirty_pages
, int mark
)
851 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
852 struct extent_state
*cached_state
= NULL
;
856 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
857 EXTENT_NEED_WAIT
, &cached_state
)) {
858 clear_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
,
859 0, 0, &cached_state
, GFP_NOFS
);
860 err
= filemap_fdatawait_range(mapping
, start
, end
);
872 * when btree blocks are allocated, they have some corresponding bits set for
873 * them in one of two extent_io trees. This is used to make sure all of
874 * those extents are on disk for transaction or log commit
876 static int btrfs_write_and_wait_marked_extents(struct btrfs_root
*root
,
877 struct extent_io_tree
*dirty_pages
, int mark
)
881 struct blk_plug plug
;
883 blk_start_plug(&plug
);
884 ret
= btrfs_write_marked_extents(root
, dirty_pages
, mark
);
885 blk_finish_plug(&plug
);
886 ret2
= btrfs_wait_marked_extents(root
, dirty_pages
, mark
);
895 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
,
896 struct btrfs_root
*root
)
898 if (!trans
|| !trans
->transaction
) {
899 struct inode
*btree_inode
;
900 btree_inode
= root
->fs_info
->btree_inode
;
901 return filemap_write_and_wait(btree_inode
->i_mapping
);
903 return btrfs_write_and_wait_marked_extents(root
,
904 &trans
->transaction
->dirty_pages
,
909 * this is used to update the root pointer in the tree of tree roots.
911 * But, in the case of the extent allocation tree, updating the root
912 * pointer may allocate blocks which may change the root of the extent
915 * So, this loops and repeats and makes sure the cowonly root didn't
916 * change while the root pointer was being updated in the metadata.
918 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
919 struct btrfs_root
*root
)
924 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
926 old_root_used
= btrfs_root_used(&root
->root_item
);
927 btrfs_write_dirty_block_groups(trans
, root
);
930 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
931 if (old_root_bytenr
== root
->node
->start
&&
932 old_root_used
== btrfs_root_used(&root
->root_item
))
935 btrfs_set_root_node(&root
->root_item
, root
->node
);
936 ret
= btrfs_update_root(trans
, tree_root
,
942 old_root_used
= btrfs_root_used(&root
->root_item
);
943 ret
= btrfs_write_dirty_block_groups(trans
, root
);
952 * update all the cowonly tree roots on disk
954 * The error handling in this function may not be obvious. Any of the
955 * failures will cause the file system to go offline. We still need
956 * to clean up the delayed refs.
958 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
,
959 struct btrfs_root
*root
)
961 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
962 struct list_head
*next
;
963 struct extent_buffer
*eb
;
966 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
970 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
971 ret
= btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
,
973 btrfs_tree_unlock(eb
);
974 free_extent_buffer(eb
);
979 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
983 ret
= btrfs_run_dev_stats(trans
, root
->fs_info
);
986 ret
= btrfs_run_dev_replace(trans
, root
->fs_info
);
989 ret
= btrfs_run_qgroups(trans
, root
->fs_info
);
993 /* run_qgroups might have added some more refs */
994 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
998 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
999 next
= fs_info
->dirty_cowonly_roots
.next
;
1000 list_del_init(next
);
1001 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
1003 if (root
!= fs_info
->extent_root
)
1004 list_add_tail(&root
->dirty_list
,
1005 &trans
->transaction
->switch_commits
);
1006 ret
= update_cowonly_root(trans
, root
);
1011 list_add_tail(&fs_info
->extent_root
->dirty_list
,
1012 &trans
->transaction
->switch_commits
);
1013 btrfs_after_dev_replace_commit(fs_info
);
1019 * dead roots are old snapshots that need to be deleted. This allocates
1020 * a dirty root struct and adds it into the list of dead roots that need to
1023 void btrfs_add_dead_root(struct btrfs_root
*root
)
1025 spin_lock(&root
->fs_info
->trans_lock
);
1026 if (list_empty(&root
->root_list
))
1027 list_add_tail(&root
->root_list
, &root
->fs_info
->dead_roots
);
1028 spin_unlock(&root
->fs_info
->trans_lock
);
1032 * update all the cowonly tree roots on disk
1034 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
,
1035 struct btrfs_root
*root
)
1037 struct btrfs_root
*gang
[8];
1038 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1043 spin_lock(&fs_info
->fs_roots_radix_lock
);
1045 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
1048 BTRFS_ROOT_TRANS_TAG
);
1051 for (i
= 0; i
< ret
; i
++) {
1053 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
1054 (unsigned long)root
->root_key
.objectid
,
1055 BTRFS_ROOT_TRANS_TAG
);
1056 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1058 btrfs_free_log(trans
, root
);
1059 btrfs_update_reloc_root(trans
, root
);
1060 btrfs_orphan_commit_root(trans
, root
);
1062 btrfs_save_ino_cache(root
, trans
);
1064 /* see comments in should_cow_block() */
1065 clear_bit(BTRFS_ROOT_FORCE_COW
, &root
->state
);
1066 smp_mb__after_atomic();
1068 if (root
->commit_root
!= root
->node
) {
1069 list_add_tail(&root
->dirty_list
,
1070 &trans
->transaction
->switch_commits
);
1071 btrfs_set_root_node(&root
->root_item
,
1075 err
= btrfs_update_root(trans
, fs_info
->tree_root
,
1078 spin_lock(&fs_info
->fs_roots_radix_lock
);
1083 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1088 * defrag a given btree.
1089 * Every leaf in the btree is read and defragged.
1091 int btrfs_defrag_root(struct btrfs_root
*root
)
1093 struct btrfs_fs_info
*info
= root
->fs_info
;
1094 struct btrfs_trans_handle
*trans
;
1097 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING
, &root
->state
))
1101 trans
= btrfs_start_transaction(root
, 0);
1103 return PTR_ERR(trans
);
1105 ret
= btrfs_defrag_leaves(trans
, root
);
1107 btrfs_end_transaction(trans
, root
);
1108 btrfs_btree_balance_dirty(info
->tree_root
);
1111 if (btrfs_fs_closing(root
->fs_info
) || ret
!= -EAGAIN
)
1114 if (btrfs_defrag_cancelled(root
->fs_info
)) {
1115 pr_debug("BTRFS: defrag_root cancelled\n");
1120 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING
, &root
->state
);
1125 * new snapshots need to be created at a very specific time in the
1126 * transaction commit. This does the actual creation.
1129 * If the error which may affect the commitment of the current transaction
1130 * happens, we should return the error number. If the error which just affect
1131 * the creation of the pending snapshots, just return 0.
1133 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
1134 struct btrfs_fs_info
*fs_info
,
1135 struct btrfs_pending_snapshot
*pending
)
1137 struct btrfs_key key
;
1138 struct btrfs_root_item
*new_root_item
;
1139 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1140 struct btrfs_root
*root
= pending
->root
;
1141 struct btrfs_root
*parent_root
;
1142 struct btrfs_block_rsv
*rsv
;
1143 struct inode
*parent_inode
;
1144 struct btrfs_path
*path
;
1145 struct btrfs_dir_item
*dir_item
;
1146 struct dentry
*dentry
;
1147 struct extent_buffer
*tmp
;
1148 struct extent_buffer
*old
;
1149 struct timespec cur_time
= CURRENT_TIME
;
1157 path
= btrfs_alloc_path();
1159 pending
->error
= -ENOMEM
;
1163 new_root_item
= kmalloc(sizeof(*new_root_item
), GFP_NOFS
);
1164 if (!new_root_item
) {
1165 pending
->error
= -ENOMEM
;
1166 goto root_item_alloc_fail
;
1169 pending
->error
= btrfs_find_free_objectid(tree_root
, &objectid
);
1171 goto no_free_objectid
;
1173 btrfs_reloc_pre_snapshot(trans
, pending
, &to_reserve
);
1175 if (to_reserve
> 0) {
1176 pending
->error
= btrfs_block_rsv_add(root
,
1177 &pending
->block_rsv
,
1179 BTRFS_RESERVE_NO_FLUSH
);
1181 goto no_free_objectid
;
1184 key
.objectid
= objectid
;
1185 key
.offset
= (u64
)-1;
1186 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1188 rsv
= trans
->block_rsv
;
1189 trans
->block_rsv
= &pending
->block_rsv
;
1190 trans
->bytes_reserved
= trans
->block_rsv
->reserved
;
1192 dentry
= pending
->dentry
;
1193 parent_inode
= pending
->dir
;
1194 parent_root
= BTRFS_I(parent_inode
)->root
;
1195 record_root_in_trans(trans
, parent_root
);
1198 * insert the directory item
1200 ret
= btrfs_set_inode_index(parent_inode
, &index
);
1201 BUG_ON(ret
); /* -ENOMEM */
1203 /* check if there is a file/dir which has the same name. */
1204 dir_item
= btrfs_lookup_dir_item(NULL
, parent_root
, path
,
1205 btrfs_ino(parent_inode
),
1206 dentry
->d_name
.name
,
1207 dentry
->d_name
.len
, 0);
1208 if (dir_item
!= NULL
&& !IS_ERR(dir_item
)) {
1209 pending
->error
= -EEXIST
;
1210 goto dir_item_existed
;
1211 } else if (IS_ERR(dir_item
)) {
1212 ret
= PTR_ERR(dir_item
);
1213 btrfs_abort_transaction(trans
, root
, ret
);
1216 btrfs_release_path(path
);
1219 * pull in the delayed directory update
1220 * and the delayed inode item
1221 * otherwise we corrupt the FS during
1224 ret
= btrfs_run_delayed_items(trans
, root
);
1225 if (ret
) { /* Transaction aborted */
1226 btrfs_abort_transaction(trans
, root
, ret
);
1230 record_root_in_trans(trans
, root
);
1231 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
1232 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
1233 btrfs_check_and_init_root_item(new_root_item
);
1235 root_flags
= btrfs_root_flags(new_root_item
);
1236 if (pending
->readonly
)
1237 root_flags
|= BTRFS_ROOT_SUBVOL_RDONLY
;
1239 root_flags
&= ~BTRFS_ROOT_SUBVOL_RDONLY
;
1240 btrfs_set_root_flags(new_root_item
, root_flags
);
1242 btrfs_set_root_generation_v2(new_root_item
,
1244 uuid_le_gen(&new_uuid
);
1245 memcpy(new_root_item
->uuid
, new_uuid
.b
, BTRFS_UUID_SIZE
);
1246 memcpy(new_root_item
->parent_uuid
, root
->root_item
.uuid
,
1248 if (!(root_flags
& BTRFS_ROOT_SUBVOL_RDONLY
)) {
1249 memset(new_root_item
->received_uuid
, 0,
1250 sizeof(new_root_item
->received_uuid
));
1251 memset(&new_root_item
->stime
, 0, sizeof(new_root_item
->stime
));
1252 memset(&new_root_item
->rtime
, 0, sizeof(new_root_item
->rtime
));
1253 btrfs_set_root_stransid(new_root_item
, 0);
1254 btrfs_set_root_rtransid(new_root_item
, 0);
1256 btrfs_set_stack_timespec_sec(&new_root_item
->otime
, cur_time
.tv_sec
);
1257 btrfs_set_stack_timespec_nsec(&new_root_item
->otime
, cur_time
.tv_nsec
);
1258 btrfs_set_root_otransid(new_root_item
, trans
->transid
);
1260 old
= btrfs_lock_root_node(root
);
1261 ret
= btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
);
1263 btrfs_tree_unlock(old
);
1264 free_extent_buffer(old
);
1265 btrfs_abort_transaction(trans
, root
, ret
);
1269 btrfs_set_lock_blocking(old
);
1271 ret
= btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
1272 /* clean up in any case */
1273 btrfs_tree_unlock(old
);
1274 free_extent_buffer(old
);
1276 btrfs_abort_transaction(trans
, root
, ret
);
1281 * We need to flush delayed refs in order to make sure all of our quota
1282 * operations have been done before we call btrfs_qgroup_inherit.
1284 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1286 btrfs_abort_transaction(trans
, root
, ret
);
1290 ret
= btrfs_qgroup_inherit(trans
, fs_info
,
1291 root
->root_key
.objectid
,
1292 objectid
, pending
->inherit
);
1294 btrfs_abort_transaction(trans
, root
, ret
);
1298 /* see comments in should_cow_block() */
1299 set_bit(BTRFS_ROOT_FORCE_COW
, &root
->state
);
1302 btrfs_set_root_node(new_root_item
, tmp
);
1303 /* record when the snapshot was created in key.offset */
1304 key
.offset
= trans
->transid
;
1305 ret
= btrfs_insert_root(trans
, tree_root
, &key
, new_root_item
);
1306 btrfs_tree_unlock(tmp
);
1307 free_extent_buffer(tmp
);
1309 btrfs_abort_transaction(trans
, root
, ret
);
1314 * insert root back/forward references
1316 ret
= btrfs_add_root_ref(trans
, tree_root
, objectid
,
1317 parent_root
->root_key
.objectid
,
1318 btrfs_ino(parent_inode
), index
,
1319 dentry
->d_name
.name
, dentry
->d_name
.len
);
1321 btrfs_abort_transaction(trans
, root
, ret
);
1325 key
.offset
= (u64
)-1;
1326 pending
->snap
= btrfs_read_fs_root_no_name(root
->fs_info
, &key
);
1327 if (IS_ERR(pending
->snap
)) {
1328 ret
= PTR_ERR(pending
->snap
);
1329 btrfs_abort_transaction(trans
, root
, ret
);
1333 ret
= btrfs_reloc_post_snapshot(trans
, pending
);
1335 btrfs_abort_transaction(trans
, root
, ret
);
1339 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1341 btrfs_abort_transaction(trans
, root
, ret
);
1345 ret
= btrfs_insert_dir_item(trans
, parent_root
,
1346 dentry
->d_name
.name
, dentry
->d_name
.len
,
1348 BTRFS_FT_DIR
, index
);
1349 /* We have check then name at the beginning, so it is impossible. */
1350 BUG_ON(ret
== -EEXIST
|| ret
== -EOVERFLOW
);
1352 btrfs_abort_transaction(trans
, root
, ret
);
1356 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
1357 dentry
->d_name
.len
* 2);
1358 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
1359 ret
= btrfs_update_inode_fallback(trans
, parent_root
, parent_inode
);
1361 btrfs_abort_transaction(trans
, root
, ret
);
1364 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
, new_uuid
.b
,
1365 BTRFS_UUID_KEY_SUBVOL
, objectid
);
1367 btrfs_abort_transaction(trans
, root
, ret
);
1370 if (!btrfs_is_empty_uuid(new_root_item
->received_uuid
)) {
1371 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
1372 new_root_item
->received_uuid
,
1373 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
1375 if (ret
&& ret
!= -EEXIST
) {
1376 btrfs_abort_transaction(trans
, root
, ret
);
1381 pending
->error
= ret
;
1383 trans
->block_rsv
= rsv
;
1384 trans
->bytes_reserved
= 0;
1386 kfree(new_root_item
);
1387 root_item_alloc_fail
:
1388 btrfs_free_path(path
);
1393 * create all the snapshots we've scheduled for creation
1395 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
,
1396 struct btrfs_fs_info
*fs_info
)
1398 struct btrfs_pending_snapshot
*pending
, *next
;
1399 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
1402 list_for_each_entry_safe(pending
, next
, head
, list
) {
1403 list_del(&pending
->list
);
1404 ret
= create_pending_snapshot(trans
, fs_info
, pending
);
1411 static void update_super_roots(struct btrfs_root
*root
)
1413 struct btrfs_root_item
*root_item
;
1414 struct btrfs_super_block
*super
;
1416 super
= root
->fs_info
->super_copy
;
1418 root_item
= &root
->fs_info
->chunk_root
->root_item
;
1419 super
->chunk_root
= root_item
->bytenr
;
1420 super
->chunk_root_generation
= root_item
->generation
;
1421 super
->chunk_root_level
= root_item
->level
;
1423 root_item
= &root
->fs_info
->tree_root
->root_item
;
1424 super
->root
= root_item
->bytenr
;
1425 super
->generation
= root_item
->generation
;
1426 super
->root_level
= root_item
->level
;
1427 if (btrfs_test_opt(root
, SPACE_CACHE
))
1428 super
->cache_generation
= root_item
->generation
;
1429 if (root
->fs_info
->update_uuid_tree_gen
)
1430 super
->uuid_tree_generation
= root_item
->generation
;
1433 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
1435 struct btrfs_transaction
*trans
;
1438 spin_lock(&info
->trans_lock
);
1439 trans
= info
->running_transaction
;
1441 ret
= (trans
->state
>= TRANS_STATE_COMMIT_START
);
1442 spin_unlock(&info
->trans_lock
);
1446 int btrfs_transaction_blocked(struct btrfs_fs_info
*info
)
1448 struct btrfs_transaction
*trans
;
1451 spin_lock(&info
->trans_lock
);
1452 trans
= info
->running_transaction
;
1454 ret
= is_transaction_blocked(trans
);
1455 spin_unlock(&info
->trans_lock
);
1460 * wait for the current transaction commit to start and block subsequent
1463 static void wait_current_trans_commit_start(struct btrfs_root
*root
,
1464 struct btrfs_transaction
*trans
)
1466 wait_event(root
->fs_info
->transaction_blocked_wait
,
1467 trans
->state
>= TRANS_STATE_COMMIT_START
||
1472 * wait for the current transaction to start and then become unblocked.
1475 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root
*root
,
1476 struct btrfs_transaction
*trans
)
1478 wait_event(root
->fs_info
->transaction_wait
,
1479 trans
->state
>= TRANS_STATE_UNBLOCKED
||
1484 * commit transactions asynchronously. once btrfs_commit_transaction_async
1485 * returns, any subsequent transaction will not be allowed to join.
1487 struct btrfs_async_commit
{
1488 struct btrfs_trans_handle
*newtrans
;
1489 struct btrfs_root
*root
;
1490 struct work_struct work
;
1493 static void do_async_commit(struct work_struct
*work
)
1495 struct btrfs_async_commit
*ac
=
1496 container_of(work
, struct btrfs_async_commit
, work
);
1499 * We've got freeze protection passed with the transaction.
1500 * Tell lockdep about it.
1502 if (ac
->newtrans
->type
& __TRANS_FREEZABLE
)
1504 &ac
->root
->fs_info
->sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
1507 current
->journal_info
= ac
->newtrans
;
1509 btrfs_commit_transaction(ac
->newtrans
, ac
->root
);
1513 int btrfs_commit_transaction_async(struct btrfs_trans_handle
*trans
,
1514 struct btrfs_root
*root
,
1515 int wait_for_unblock
)
1517 struct btrfs_async_commit
*ac
;
1518 struct btrfs_transaction
*cur_trans
;
1520 ac
= kmalloc(sizeof(*ac
), GFP_NOFS
);
1524 INIT_WORK(&ac
->work
, do_async_commit
);
1526 ac
->newtrans
= btrfs_join_transaction(root
);
1527 if (IS_ERR(ac
->newtrans
)) {
1528 int err
= PTR_ERR(ac
->newtrans
);
1533 /* take transaction reference */
1534 cur_trans
= trans
->transaction
;
1535 atomic_inc(&cur_trans
->use_count
);
1537 btrfs_end_transaction(trans
, root
);
1540 * Tell lockdep we've released the freeze rwsem, since the
1541 * async commit thread will be the one to unlock it.
1543 if (ac
->newtrans
->type
& __TRANS_FREEZABLE
)
1545 &root
->fs_info
->sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
1548 schedule_work(&ac
->work
);
1550 /* wait for transaction to start and unblock */
1551 if (wait_for_unblock
)
1552 wait_current_trans_commit_start_and_unblock(root
, cur_trans
);
1554 wait_current_trans_commit_start(root
, cur_trans
);
1556 if (current
->journal_info
== trans
)
1557 current
->journal_info
= NULL
;
1559 btrfs_put_transaction(cur_trans
);
1564 static void cleanup_transaction(struct btrfs_trans_handle
*trans
,
1565 struct btrfs_root
*root
, int err
)
1567 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1570 WARN_ON(trans
->use_count
> 1);
1572 btrfs_abort_transaction(trans
, root
, err
);
1574 spin_lock(&root
->fs_info
->trans_lock
);
1577 * If the transaction is removed from the list, it means this
1578 * transaction has been committed successfully, so it is impossible
1579 * to call the cleanup function.
1581 BUG_ON(list_empty(&cur_trans
->list
));
1583 list_del_init(&cur_trans
->list
);
1584 if (cur_trans
== root
->fs_info
->running_transaction
) {
1585 cur_trans
->state
= TRANS_STATE_COMMIT_DOING
;
1586 spin_unlock(&root
->fs_info
->trans_lock
);
1587 wait_event(cur_trans
->writer_wait
,
1588 atomic_read(&cur_trans
->num_writers
) == 1);
1590 spin_lock(&root
->fs_info
->trans_lock
);
1592 spin_unlock(&root
->fs_info
->trans_lock
);
1594 btrfs_cleanup_one_transaction(trans
->transaction
, root
);
1596 spin_lock(&root
->fs_info
->trans_lock
);
1597 if (cur_trans
== root
->fs_info
->running_transaction
)
1598 root
->fs_info
->running_transaction
= NULL
;
1599 spin_unlock(&root
->fs_info
->trans_lock
);
1601 if (trans
->type
& __TRANS_FREEZABLE
)
1602 sb_end_intwrite(root
->fs_info
->sb
);
1603 btrfs_put_transaction(cur_trans
);
1604 btrfs_put_transaction(cur_trans
);
1606 trace_btrfs_transaction_commit(root
);
1608 if (current
->journal_info
== trans
)
1609 current
->journal_info
= NULL
;
1610 btrfs_scrub_cancel(root
->fs_info
);
1612 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1615 static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle
*trans
,
1616 struct btrfs_root
*root
)
1620 ret
= btrfs_run_delayed_items(trans
, root
);
1625 * rename don't use btrfs_join_transaction, so, once we
1626 * set the transaction to blocked above, we aren't going
1627 * to get any new ordered operations. We can safely run
1628 * it here and no for sure that nothing new will be added
1631 ret
= btrfs_run_ordered_operations(trans
, root
, 1);
1636 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info
*fs_info
)
1638 if (btrfs_test_opt(fs_info
->tree_root
, FLUSHONCOMMIT
))
1639 return btrfs_start_delalloc_roots(fs_info
, 1, -1);
1643 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info
*fs_info
)
1645 if (btrfs_test_opt(fs_info
->tree_root
, FLUSHONCOMMIT
))
1646 btrfs_wait_ordered_roots(fs_info
, -1);
1649 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
,
1650 struct btrfs_root
*root
)
1652 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1653 struct btrfs_transaction
*prev_trans
= NULL
;
1656 ret
= btrfs_run_ordered_operations(trans
, root
, 0);
1658 btrfs_abort_transaction(trans
, root
, ret
);
1659 btrfs_end_transaction(trans
, root
);
1663 /* Stop the commit early if ->aborted is set */
1664 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1665 ret
= cur_trans
->aborted
;
1666 btrfs_end_transaction(trans
, root
);
1670 /* make a pass through all the delayed refs we have so far
1671 * any runnings procs may add more while we are here
1673 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1675 btrfs_end_transaction(trans
, root
);
1679 btrfs_trans_release_metadata(trans
, root
);
1680 trans
->block_rsv
= NULL
;
1681 if (trans
->qgroup_reserved
) {
1682 btrfs_qgroup_free(root
, trans
->qgroup_reserved
);
1683 trans
->qgroup_reserved
= 0;
1686 cur_trans
= trans
->transaction
;
1689 * set the flushing flag so procs in this transaction have to
1690 * start sending their work down.
1692 cur_trans
->delayed_refs
.flushing
= 1;
1695 if (!list_empty(&trans
->new_bgs
))
1696 btrfs_create_pending_block_groups(trans
, root
);
1698 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1700 btrfs_end_transaction(trans
, root
);
1704 spin_lock(&root
->fs_info
->trans_lock
);
1705 if (cur_trans
->state
>= TRANS_STATE_COMMIT_START
) {
1706 spin_unlock(&root
->fs_info
->trans_lock
);
1707 atomic_inc(&cur_trans
->use_count
);
1708 ret
= btrfs_end_transaction(trans
, root
);
1710 wait_for_commit(root
, cur_trans
);
1712 btrfs_put_transaction(cur_trans
);
1717 cur_trans
->state
= TRANS_STATE_COMMIT_START
;
1718 wake_up(&root
->fs_info
->transaction_blocked_wait
);
1720 if (cur_trans
->list
.prev
!= &root
->fs_info
->trans_list
) {
1721 prev_trans
= list_entry(cur_trans
->list
.prev
,
1722 struct btrfs_transaction
, list
);
1723 if (prev_trans
->state
!= TRANS_STATE_COMPLETED
) {
1724 atomic_inc(&prev_trans
->use_count
);
1725 spin_unlock(&root
->fs_info
->trans_lock
);
1727 wait_for_commit(root
, prev_trans
);
1729 btrfs_put_transaction(prev_trans
);
1731 spin_unlock(&root
->fs_info
->trans_lock
);
1734 spin_unlock(&root
->fs_info
->trans_lock
);
1737 extwriter_counter_dec(cur_trans
, trans
->type
);
1739 ret
= btrfs_start_delalloc_flush(root
->fs_info
);
1741 goto cleanup_transaction
;
1743 ret
= btrfs_flush_all_pending_stuffs(trans
, root
);
1745 goto cleanup_transaction
;
1747 wait_event(cur_trans
->writer_wait
,
1748 extwriter_counter_read(cur_trans
) == 0);
1750 /* some pending stuffs might be added after the previous flush. */
1751 ret
= btrfs_flush_all_pending_stuffs(trans
, root
);
1753 goto cleanup_transaction
;
1755 btrfs_wait_delalloc_flush(root
->fs_info
);
1757 btrfs_scrub_pause(root
);
1759 * Ok now we need to make sure to block out any other joins while we
1760 * commit the transaction. We could have started a join before setting
1761 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
1763 spin_lock(&root
->fs_info
->trans_lock
);
1764 cur_trans
->state
= TRANS_STATE_COMMIT_DOING
;
1765 spin_unlock(&root
->fs_info
->trans_lock
);
1766 wait_event(cur_trans
->writer_wait
,
1767 atomic_read(&cur_trans
->num_writers
) == 1);
1769 /* ->aborted might be set after the previous check, so check it */
1770 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1771 ret
= cur_trans
->aborted
;
1772 goto scrub_continue
;
1775 * the reloc mutex makes sure that we stop
1776 * the balancing code from coming in and moving
1777 * extents around in the middle of the commit
1779 mutex_lock(&root
->fs_info
->reloc_mutex
);
1782 * We needn't worry about the delayed items because we will
1783 * deal with them in create_pending_snapshot(), which is the
1784 * core function of the snapshot creation.
1786 ret
= create_pending_snapshots(trans
, root
->fs_info
);
1788 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1789 goto scrub_continue
;
1793 * We insert the dir indexes of the snapshots and update the inode
1794 * of the snapshots' parents after the snapshot creation, so there
1795 * are some delayed items which are not dealt with. Now deal with
1798 * We needn't worry that this operation will corrupt the snapshots,
1799 * because all the tree which are snapshoted will be forced to COW
1800 * the nodes and leaves.
1802 ret
= btrfs_run_delayed_items(trans
, root
);
1804 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1805 goto scrub_continue
;
1808 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1810 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1811 goto scrub_continue
;
1815 * make sure none of the code above managed to slip in a
1818 btrfs_assert_delayed_root_empty(root
);
1820 WARN_ON(cur_trans
!= trans
->transaction
);
1822 /* btrfs_commit_tree_roots is responsible for getting the
1823 * various roots consistent with each other. Every pointer
1824 * in the tree of tree roots has to point to the most up to date
1825 * root for every subvolume and other tree. So, we have to keep
1826 * the tree logging code from jumping in and changing any
1829 * At this point in the commit, there can't be any tree-log
1830 * writers, but a little lower down we drop the trans mutex
1831 * and let new people in. By holding the tree_log_mutex
1832 * from now until after the super is written, we avoid races
1833 * with the tree-log code.
1835 mutex_lock(&root
->fs_info
->tree_log_mutex
);
1837 ret
= commit_fs_roots(trans
, root
);
1839 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1840 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1841 goto scrub_continue
;
1845 * Since the transaction is done, we should set the inode map cache flag
1846 * before any other comming transaction.
1848 if (btrfs_test_opt(root
, CHANGE_INODE_CACHE
))
1849 btrfs_set_opt(root
->fs_info
->mount_opt
, INODE_MAP_CACHE
);
1851 btrfs_clear_opt(root
->fs_info
->mount_opt
, INODE_MAP_CACHE
);
1853 /* commit_fs_roots gets rid of all the tree log roots, it is now
1854 * safe to free the root of tree log roots
1856 btrfs_free_log_root_tree(trans
, root
->fs_info
);
1858 ret
= commit_cowonly_roots(trans
, root
);
1860 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1861 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1862 goto scrub_continue
;
1866 * The tasks which save the space cache and inode cache may also
1867 * update ->aborted, check it.
1869 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1870 ret
= cur_trans
->aborted
;
1871 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1872 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1873 goto scrub_continue
;
1876 btrfs_prepare_extent_commit(trans
, root
);
1878 cur_trans
= root
->fs_info
->running_transaction
;
1880 btrfs_set_root_node(&root
->fs_info
->tree_root
->root_item
,
1881 root
->fs_info
->tree_root
->node
);
1882 list_add_tail(&root
->fs_info
->tree_root
->dirty_list
,
1883 &cur_trans
->switch_commits
);
1885 btrfs_set_root_node(&root
->fs_info
->chunk_root
->root_item
,
1886 root
->fs_info
->chunk_root
->node
);
1887 list_add_tail(&root
->fs_info
->chunk_root
->dirty_list
,
1888 &cur_trans
->switch_commits
);
1890 switch_commit_roots(cur_trans
, root
->fs_info
);
1892 assert_qgroups_uptodate(trans
);
1893 update_super_roots(root
);
1895 btrfs_set_super_log_root(root
->fs_info
->super_copy
, 0);
1896 btrfs_set_super_log_root_level(root
->fs_info
->super_copy
, 0);
1897 memcpy(root
->fs_info
->super_for_commit
, root
->fs_info
->super_copy
,
1898 sizeof(*root
->fs_info
->super_copy
));
1900 spin_lock(&root
->fs_info
->trans_lock
);
1901 cur_trans
->state
= TRANS_STATE_UNBLOCKED
;
1902 root
->fs_info
->running_transaction
= NULL
;
1903 spin_unlock(&root
->fs_info
->trans_lock
);
1904 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1906 wake_up(&root
->fs_info
->transaction_wait
);
1908 ret
= btrfs_write_and_wait_transaction(trans
, root
);
1910 btrfs_error(root
->fs_info
, ret
,
1911 "Error while writing out transaction");
1912 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1913 goto scrub_continue
;
1916 ret
= write_ctree_super(trans
, root
, 0);
1918 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1919 goto scrub_continue
;
1923 * the super is written, we can safely allow the tree-loggers
1924 * to go about their business
1926 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1928 btrfs_finish_extent_commit(trans
, root
);
1930 root
->fs_info
->last_trans_committed
= cur_trans
->transid
;
1932 * We needn't acquire the lock here because there is no other task
1933 * which can change it.
1935 cur_trans
->state
= TRANS_STATE_COMPLETED
;
1936 wake_up(&cur_trans
->commit_wait
);
1938 spin_lock(&root
->fs_info
->trans_lock
);
1939 list_del_init(&cur_trans
->list
);
1940 spin_unlock(&root
->fs_info
->trans_lock
);
1942 btrfs_put_transaction(cur_trans
);
1943 btrfs_put_transaction(cur_trans
);
1945 if (trans
->type
& __TRANS_FREEZABLE
)
1946 sb_end_intwrite(root
->fs_info
->sb
);
1948 trace_btrfs_transaction_commit(root
);
1950 btrfs_scrub_continue(root
);
1952 if (current
->journal_info
== trans
)
1953 current
->journal_info
= NULL
;
1955 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1957 if (current
!= root
->fs_info
->transaction_kthread
)
1958 btrfs_run_delayed_iputs(root
);
1963 btrfs_scrub_continue(root
);
1964 cleanup_transaction
:
1965 btrfs_trans_release_metadata(trans
, root
);
1966 trans
->block_rsv
= NULL
;
1967 if (trans
->qgroup_reserved
) {
1968 btrfs_qgroup_free(root
, trans
->qgroup_reserved
);
1969 trans
->qgroup_reserved
= 0;
1971 btrfs_warn(root
->fs_info
, "Skipping commit of aborted transaction.");
1972 if (current
->journal_info
== trans
)
1973 current
->journal_info
= NULL
;
1974 cleanup_transaction(trans
, root
, ret
);
1980 * return < 0 if error
1981 * 0 if there are no more dead_roots at the time of call
1982 * 1 there are more to be processed, call me again
1984 * The return value indicates there are certainly more snapshots to delete, but
1985 * if there comes a new one during processing, it may return 0. We don't mind,
1986 * because btrfs_commit_super will poke cleaner thread and it will process it a
1987 * few seconds later.
1989 int btrfs_clean_one_deleted_snapshot(struct btrfs_root
*root
)
1992 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1994 spin_lock(&fs_info
->trans_lock
);
1995 if (list_empty(&fs_info
->dead_roots
)) {
1996 spin_unlock(&fs_info
->trans_lock
);
1999 root
= list_first_entry(&fs_info
->dead_roots
,
2000 struct btrfs_root
, root_list
);
2001 list_del_init(&root
->root_list
);
2002 spin_unlock(&fs_info
->trans_lock
);
2004 pr_debug("BTRFS: cleaner removing %llu\n", root
->objectid
);
2006 btrfs_kill_all_delayed_nodes(root
);
2008 if (btrfs_header_backref_rev(root
->node
) <
2009 BTRFS_MIXED_BACKREF_REV
)
2010 ret
= btrfs_drop_snapshot(root
, NULL
, 0, 0);
2012 ret
= btrfs_drop_snapshot(root
, NULL
, 1, 0);
2014 * If we encounter a transaction abort during snapshot cleaning, we
2015 * don't want to crash here
2017 return (ret
< 0) ? 0 : 1;