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 void clear_btree_io_tree(struct extent_io_tree
*tree
)
81 spin_lock(&tree
->lock
);
82 while (!RB_EMPTY_ROOT(&tree
->state
)) {
84 struct extent_state
*state
;
86 node
= rb_first(&tree
->state
);
87 state
= rb_entry(node
, struct extent_state
, rb_node
);
88 rb_erase(&state
->rb_node
, &tree
->state
);
89 RB_CLEAR_NODE(&state
->rb_node
);
91 * btree io trees aren't supposed to have tasks waiting for
92 * changes in the flags of extent states ever.
94 ASSERT(!waitqueue_active(&state
->wq
));
95 free_extent_state(state
);
97 spin_unlock(&tree
->lock
);
99 spin_lock(&tree
->lock
);
102 spin_unlock(&tree
->lock
);
105 static noinline
void switch_commit_roots(struct btrfs_transaction
*trans
,
106 struct btrfs_fs_info
*fs_info
)
108 struct btrfs_root
*root
, *tmp
;
110 down_write(&fs_info
->commit_root_sem
);
111 list_for_each_entry_safe(root
, tmp
, &trans
->switch_commits
,
113 list_del_init(&root
->dirty_list
);
114 free_extent_buffer(root
->commit_root
);
115 root
->commit_root
= btrfs_root_node(root
);
116 if (is_fstree(root
->objectid
))
117 btrfs_unpin_free_ino(root
);
118 clear_btree_io_tree(&root
->dirty_log_pages
);
120 up_write(&fs_info
->commit_root_sem
);
123 static inline void extwriter_counter_inc(struct btrfs_transaction
*trans
,
126 if (type
& TRANS_EXTWRITERS
)
127 atomic_inc(&trans
->num_extwriters
);
130 static inline void extwriter_counter_dec(struct btrfs_transaction
*trans
,
133 if (type
& TRANS_EXTWRITERS
)
134 atomic_dec(&trans
->num_extwriters
);
137 static inline void extwriter_counter_init(struct btrfs_transaction
*trans
,
140 atomic_set(&trans
->num_extwriters
, ((type
& TRANS_EXTWRITERS
) ? 1 : 0));
143 static inline int extwriter_counter_read(struct btrfs_transaction
*trans
)
145 return atomic_read(&trans
->num_extwriters
);
149 * either allocate a new transaction or hop into the existing one
151 static noinline
int join_transaction(struct btrfs_root
*root
, unsigned int type
)
153 struct btrfs_transaction
*cur_trans
;
154 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
156 spin_lock(&fs_info
->trans_lock
);
158 /* The file system has been taken offline. No new transactions. */
159 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
160 spin_unlock(&fs_info
->trans_lock
);
164 cur_trans
= fs_info
->running_transaction
;
166 if (cur_trans
->aborted
) {
167 spin_unlock(&fs_info
->trans_lock
);
168 return cur_trans
->aborted
;
170 if (btrfs_blocked_trans_types
[cur_trans
->state
] & type
) {
171 spin_unlock(&fs_info
->trans_lock
);
174 atomic_inc(&cur_trans
->use_count
);
175 atomic_inc(&cur_trans
->num_writers
);
176 extwriter_counter_inc(cur_trans
, type
);
177 spin_unlock(&fs_info
->trans_lock
);
180 spin_unlock(&fs_info
->trans_lock
);
183 * If we are ATTACH, we just want to catch the current transaction,
184 * and commit it. If there is no transaction, just return ENOENT.
186 if (type
== TRANS_ATTACH
)
190 * JOIN_NOLOCK only happens during the transaction commit, so
191 * it is impossible that ->running_transaction is NULL
193 BUG_ON(type
== TRANS_JOIN_NOLOCK
);
195 cur_trans
= kmem_cache_alloc(btrfs_transaction_cachep
, GFP_NOFS
);
199 spin_lock(&fs_info
->trans_lock
);
200 if (fs_info
->running_transaction
) {
202 * someone started a transaction after we unlocked. Make sure
203 * to redo the checks above
205 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
207 } else if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
208 spin_unlock(&fs_info
->trans_lock
);
209 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
213 atomic_set(&cur_trans
->num_writers
, 1);
214 extwriter_counter_init(cur_trans
, type
);
215 init_waitqueue_head(&cur_trans
->writer_wait
);
216 init_waitqueue_head(&cur_trans
->commit_wait
);
217 cur_trans
->state
= TRANS_STATE_RUNNING
;
219 * One for this trans handle, one so it will live on until we
220 * commit the transaction.
222 atomic_set(&cur_trans
->use_count
, 2);
223 cur_trans
->have_free_bgs
= 0;
224 cur_trans
->start_time
= get_seconds();
226 cur_trans
->delayed_refs
.href_root
= RB_ROOT
;
227 atomic_set(&cur_trans
->delayed_refs
.num_entries
, 0);
228 cur_trans
->delayed_refs
.num_heads_ready
= 0;
229 cur_trans
->delayed_refs
.num_heads
= 0;
230 cur_trans
->delayed_refs
.flushing
= 0;
231 cur_trans
->delayed_refs
.run_delayed_start
= 0;
234 * although the tree mod log is per file system and not per transaction,
235 * the log must never go across transaction boundaries.
238 if (!list_empty(&fs_info
->tree_mod_seq_list
))
239 WARN(1, KERN_ERR
"BTRFS: tree_mod_seq_list not empty when "
240 "creating a fresh transaction\n");
241 if (!RB_EMPTY_ROOT(&fs_info
->tree_mod_log
))
242 WARN(1, KERN_ERR
"BTRFS: tree_mod_log rb tree not empty when "
243 "creating a fresh transaction\n");
244 atomic64_set(&fs_info
->tree_mod_seq
, 0);
246 spin_lock_init(&cur_trans
->delayed_refs
.lock
);
248 INIT_LIST_HEAD(&cur_trans
->pending_snapshots
);
249 INIT_LIST_HEAD(&cur_trans
->pending_chunks
);
250 INIT_LIST_HEAD(&cur_trans
->switch_commits
);
251 INIT_LIST_HEAD(&cur_trans
->pending_ordered
);
252 INIT_LIST_HEAD(&cur_trans
->dirty_bgs
);
253 spin_lock_init(&cur_trans
->dirty_bgs_lock
);
254 list_add_tail(&cur_trans
->list
, &fs_info
->trans_list
);
255 extent_io_tree_init(&cur_trans
->dirty_pages
,
256 fs_info
->btree_inode
->i_mapping
);
257 fs_info
->generation
++;
258 cur_trans
->transid
= fs_info
->generation
;
259 fs_info
->running_transaction
= cur_trans
;
260 cur_trans
->aborted
= 0;
261 spin_unlock(&fs_info
->trans_lock
);
267 * this does all the record keeping required to make sure that a reference
268 * counted root is properly recorded in a given transaction. This is required
269 * to make sure the old root from before we joined the transaction is deleted
270 * when the transaction commits
272 static int record_root_in_trans(struct btrfs_trans_handle
*trans
,
273 struct btrfs_root
*root
)
275 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) &&
276 root
->last_trans
< trans
->transid
) {
277 WARN_ON(root
== root
->fs_info
->extent_root
);
278 WARN_ON(root
->commit_root
!= root
->node
);
281 * see below for IN_TRANS_SETUP usage rules
282 * we have the reloc mutex held now, so there
283 * is only one writer in this function
285 set_bit(BTRFS_ROOT_IN_TRANS_SETUP
, &root
->state
);
287 /* make sure readers find IN_TRANS_SETUP before
288 * they find our root->last_trans update
292 spin_lock(&root
->fs_info
->fs_roots_radix_lock
);
293 if (root
->last_trans
== trans
->transid
) {
294 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
297 radix_tree_tag_set(&root
->fs_info
->fs_roots_radix
,
298 (unsigned long)root
->root_key
.objectid
,
299 BTRFS_ROOT_TRANS_TAG
);
300 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
301 root
->last_trans
= trans
->transid
;
303 /* this is pretty tricky. We don't want to
304 * take the relocation lock in btrfs_record_root_in_trans
305 * unless we're really doing the first setup for this root in
308 * Normally we'd use root->last_trans as a flag to decide
309 * if we want to take the expensive mutex.
311 * But, we have to set root->last_trans before we
312 * init the relocation root, otherwise, we trip over warnings
313 * in ctree.c. The solution used here is to flag ourselves
314 * with root IN_TRANS_SETUP. When this is 1, we're still
315 * fixing up the reloc trees and everyone must wait.
317 * When this is zero, they can trust root->last_trans and fly
318 * through btrfs_record_root_in_trans without having to take the
319 * lock. smp_wmb() makes sure that all the writes above are
320 * done before we pop in the zero below
322 btrfs_init_reloc_root(trans
, root
);
323 smp_mb__before_atomic();
324 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP
, &root
->state
);
330 int btrfs_record_root_in_trans(struct btrfs_trans_handle
*trans
,
331 struct btrfs_root
*root
)
333 if (!test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
337 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
341 if (root
->last_trans
== trans
->transid
&&
342 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP
, &root
->state
))
345 mutex_lock(&root
->fs_info
->reloc_mutex
);
346 record_root_in_trans(trans
, root
);
347 mutex_unlock(&root
->fs_info
->reloc_mutex
);
352 static inline int is_transaction_blocked(struct btrfs_transaction
*trans
)
354 return (trans
->state
>= TRANS_STATE_BLOCKED
&&
355 trans
->state
< TRANS_STATE_UNBLOCKED
&&
359 /* wait for commit against the current transaction to become unblocked
360 * when this is done, it is safe to start a new transaction, but the current
361 * transaction might not be fully on disk.
363 static void wait_current_trans(struct btrfs_root
*root
)
365 struct btrfs_transaction
*cur_trans
;
367 spin_lock(&root
->fs_info
->trans_lock
);
368 cur_trans
= root
->fs_info
->running_transaction
;
369 if (cur_trans
&& is_transaction_blocked(cur_trans
)) {
370 atomic_inc(&cur_trans
->use_count
);
371 spin_unlock(&root
->fs_info
->trans_lock
);
373 wait_event(root
->fs_info
->transaction_wait
,
374 cur_trans
->state
>= TRANS_STATE_UNBLOCKED
||
376 btrfs_put_transaction(cur_trans
);
378 spin_unlock(&root
->fs_info
->trans_lock
);
382 static int may_wait_transaction(struct btrfs_root
*root
, int type
)
384 if (root
->fs_info
->log_root_recovering
)
387 if (type
== TRANS_USERSPACE
)
390 if (type
== TRANS_START
&&
391 !atomic_read(&root
->fs_info
->open_ioctl_trans
))
397 static inline bool need_reserve_reloc_root(struct btrfs_root
*root
)
399 if (!root
->fs_info
->reloc_ctl
||
400 !test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
401 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
408 static struct btrfs_trans_handle
*
409 start_transaction(struct btrfs_root
*root
, u64 num_items
, unsigned int type
,
410 enum btrfs_reserve_flush_enum flush
)
412 struct btrfs_trans_handle
*h
;
413 struct btrfs_transaction
*cur_trans
;
415 u64 qgroup_reserved
= 0;
416 bool reloc_reserved
= false;
419 /* Send isn't supposed to start transactions. */
420 ASSERT(current
->journal_info
!= BTRFS_SEND_TRANS_STUB
);
422 if (test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
423 return ERR_PTR(-EROFS
);
425 if (current
->journal_info
) {
426 WARN_ON(type
& TRANS_EXTWRITERS
);
427 h
= current
->journal_info
;
429 WARN_ON(h
->use_count
> 2);
430 h
->orig_rsv
= h
->block_rsv
;
436 * Do the reservation before we join the transaction so we can do all
437 * the appropriate flushing if need be.
439 if (num_items
> 0 && root
!= root
->fs_info
->chunk_root
) {
440 if (root
->fs_info
->quota_enabled
&&
441 is_fstree(root
->root_key
.objectid
)) {
442 qgroup_reserved
= num_items
* root
->nodesize
;
443 ret
= btrfs_qgroup_reserve(root
, qgroup_reserved
);
448 num_bytes
= btrfs_calc_trans_metadata_size(root
, num_items
);
450 * Do the reservation for the relocation root creation
452 if (need_reserve_reloc_root(root
)) {
453 num_bytes
+= root
->nodesize
;
454 reloc_reserved
= true;
457 ret
= btrfs_block_rsv_add(root
,
458 &root
->fs_info
->trans_block_rsv
,
464 h
= kmem_cache_alloc(btrfs_trans_handle_cachep
, GFP_NOFS
);
471 * If we are JOIN_NOLOCK we're already committing a transaction and
472 * waiting on this guy, so we don't need to do the sb_start_intwrite
473 * because we're already holding a ref. We need this because we could
474 * have raced in and did an fsync() on a file which can kick a commit
475 * and then we deadlock with somebody doing a freeze.
477 * If we are ATTACH, it means we just want to catch the current
478 * transaction and commit it, so we needn't do sb_start_intwrite().
480 if (type
& __TRANS_FREEZABLE
)
481 sb_start_intwrite(root
->fs_info
->sb
);
483 if (may_wait_transaction(root
, type
))
484 wait_current_trans(root
);
487 ret
= join_transaction(root
, type
);
489 wait_current_trans(root
);
490 if (unlikely(type
== TRANS_ATTACH
))
493 } while (ret
== -EBUSY
);
496 /* We must get the transaction if we are JOIN_NOLOCK. */
497 BUG_ON(type
== TRANS_JOIN_NOLOCK
);
501 cur_trans
= root
->fs_info
->running_transaction
;
503 h
->transid
= cur_trans
->transid
;
504 h
->transaction
= cur_trans
;
506 h
->bytes_reserved
= 0;
508 h
->delayed_ref_updates
= 0;
514 h
->qgroup_reserved
= 0;
515 h
->delayed_ref_elem
.seq
= 0;
517 h
->allocating_chunk
= false;
518 h
->reloc_reserved
= false;
520 INIT_LIST_HEAD(&h
->qgroup_ref_list
);
521 INIT_LIST_HEAD(&h
->new_bgs
);
522 INIT_LIST_HEAD(&h
->ordered
);
525 if (cur_trans
->state
>= TRANS_STATE_BLOCKED
&&
526 may_wait_transaction(root
, type
)) {
527 current
->journal_info
= h
;
528 btrfs_commit_transaction(h
, root
);
533 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
534 h
->transid
, num_bytes
, 1);
535 h
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
536 h
->bytes_reserved
= num_bytes
;
537 h
->reloc_reserved
= reloc_reserved
;
539 h
->qgroup_reserved
= qgroup_reserved
;
542 btrfs_record_root_in_trans(h
, root
);
544 if (!current
->journal_info
&& type
!= TRANS_USERSPACE
)
545 current
->journal_info
= h
;
549 if (type
& __TRANS_FREEZABLE
)
550 sb_end_intwrite(root
->fs_info
->sb
);
551 kmem_cache_free(btrfs_trans_handle_cachep
, h
);
554 btrfs_block_rsv_release(root
, &root
->fs_info
->trans_block_rsv
,
558 btrfs_qgroup_free(root
, qgroup_reserved
);
562 struct btrfs_trans_handle
*btrfs_start_transaction(struct btrfs_root
*root
,
565 return start_transaction(root
, num_items
, TRANS_START
,
566 BTRFS_RESERVE_FLUSH_ALL
);
569 struct btrfs_trans_handle
*btrfs_start_transaction_lflush(
570 struct btrfs_root
*root
, int num_items
)
572 return start_transaction(root
, num_items
, TRANS_START
,
573 BTRFS_RESERVE_FLUSH_LIMIT
);
576 struct btrfs_trans_handle
*btrfs_join_transaction(struct btrfs_root
*root
)
578 return start_transaction(root
, 0, TRANS_JOIN
, 0);
581 struct btrfs_trans_handle
*btrfs_join_transaction_nolock(struct btrfs_root
*root
)
583 return start_transaction(root
, 0, TRANS_JOIN_NOLOCK
, 0);
586 struct btrfs_trans_handle
*btrfs_start_ioctl_transaction(struct btrfs_root
*root
)
588 return start_transaction(root
, 0, TRANS_USERSPACE
, 0);
592 * btrfs_attach_transaction() - catch the running transaction
594 * It is used when we want to commit the current the transaction, but
595 * don't want to start a new one.
597 * Note: If this function return -ENOENT, it just means there is no
598 * running transaction. But it is possible that the inactive transaction
599 * is still in the memory, not fully on disk. If you hope there is no
600 * inactive transaction in the fs when -ENOENT is returned, you should
602 * btrfs_attach_transaction_barrier()
604 struct btrfs_trans_handle
*btrfs_attach_transaction(struct btrfs_root
*root
)
606 return start_transaction(root
, 0, TRANS_ATTACH
, 0);
610 * btrfs_attach_transaction_barrier() - catch the running transaction
612 * It is similar to the above function, the differentia is this one
613 * will wait for all the inactive transactions until they fully
616 struct btrfs_trans_handle
*
617 btrfs_attach_transaction_barrier(struct btrfs_root
*root
)
619 struct btrfs_trans_handle
*trans
;
621 trans
= start_transaction(root
, 0, TRANS_ATTACH
, 0);
622 if (IS_ERR(trans
) && PTR_ERR(trans
) == -ENOENT
)
623 btrfs_wait_for_commit(root
, 0);
628 /* wait for a transaction commit to be fully complete */
629 static noinline
void wait_for_commit(struct btrfs_root
*root
,
630 struct btrfs_transaction
*commit
)
632 wait_event(commit
->commit_wait
, commit
->state
== TRANS_STATE_COMPLETED
);
635 int btrfs_wait_for_commit(struct btrfs_root
*root
, u64 transid
)
637 struct btrfs_transaction
*cur_trans
= NULL
, *t
;
641 if (transid
<= root
->fs_info
->last_trans_committed
)
644 /* find specified transaction */
645 spin_lock(&root
->fs_info
->trans_lock
);
646 list_for_each_entry(t
, &root
->fs_info
->trans_list
, list
) {
647 if (t
->transid
== transid
) {
649 atomic_inc(&cur_trans
->use_count
);
653 if (t
->transid
> transid
) {
658 spin_unlock(&root
->fs_info
->trans_lock
);
661 * The specified transaction doesn't exist, or we
662 * raced with btrfs_commit_transaction
665 if (transid
> root
->fs_info
->last_trans_committed
)
670 /* find newest transaction that is committing | committed */
671 spin_lock(&root
->fs_info
->trans_lock
);
672 list_for_each_entry_reverse(t
, &root
->fs_info
->trans_list
,
674 if (t
->state
>= TRANS_STATE_COMMIT_START
) {
675 if (t
->state
== TRANS_STATE_COMPLETED
)
678 atomic_inc(&cur_trans
->use_count
);
682 spin_unlock(&root
->fs_info
->trans_lock
);
684 goto out
; /* nothing committing|committed */
687 wait_for_commit(root
, cur_trans
);
688 btrfs_put_transaction(cur_trans
);
693 void btrfs_throttle(struct btrfs_root
*root
)
695 if (!atomic_read(&root
->fs_info
->open_ioctl_trans
))
696 wait_current_trans(root
);
699 static int should_end_transaction(struct btrfs_trans_handle
*trans
,
700 struct btrfs_root
*root
)
702 if (root
->fs_info
->global_block_rsv
.space_info
->full
&&
703 btrfs_check_space_for_delayed_refs(trans
, root
))
706 return !!btrfs_block_rsv_check(root
, &root
->fs_info
->global_block_rsv
, 5);
709 int btrfs_should_end_transaction(struct btrfs_trans_handle
*trans
,
710 struct btrfs_root
*root
)
712 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
717 if (cur_trans
->state
>= TRANS_STATE_BLOCKED
||
718 cur_trans
->delayed_refs
.flushing
)
721 updates
= trans
->delayed_ref_updates
;
722 trans
->delayed_ref_updates
= 0;
724 err
= btrfs_run_delayed_refs(trans
, root
, updates
);
725 if (err
) /* Error code will also eval true */
729 return should_end_transaction(trans
, root
);
732 static int __btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
733 struct btrfs_root
*root
, int throttle
)
735 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
736 struct btrfs_fs_info
*info
= root
->fs_info
;
737 unsigned long cur
= trans
->delayed_ref_updates
;
738 int lock
= (trans
->type
!= TRANS_JOIN_NOLOCK
);
740 int must_run_delayed_refs
= 0;
742 if (trans
->use_count
> 1) {
744 trans
->block_rsv
= trans
->orig_rsv
;
748 btrfs_trans_release_metadata(trans
, root
);
749 trans
->block_rsv
= NULL
;
751 if (!list_empty(&trans
->new_bgs
))
752 btrfs_create_pending_block_groups(trans
, root
);
754 if (!list_empty(&trans
->ordered
)) {
755 spin_lock(&info
->trans_lock
);
756 list_splice(&trans
->ordered
, &cur_trans
->pending_ordered
);
757 spin_unlock(&info
->trans_lock
);
760 trans
->delayed_ref_updates
= 0;
762 must_run_delayed_refs
=
763 btrfs_should_throttle_delayed_refs(trans
, root
);
764 cur
= max_t(unsigned long, cur
, 32);
767 * don't make the caller wait if they are from a NOLOCK
768 * or ATTACH transaction, it will deadlock with commit
770 if (must_run_delayed_refs
== 1 &&
771 (trans
->type
& (__TRANS_JOIN_NOLOCK
| __TRANS_ATTACH
)))
772 must_run_delayed_refs
= 2;
775 if (trans
->qgroup_reserved
) {
777 * the same root has to be passed here between start_transaction
778 * and end_transaction. Subvolume quota depends on this.
780 btrfs_qgroup_free(trans
->root
, trans
->qgroup_reserved
);
781 trans
->qgroup_reserved
= 0;
784 btrfs_trans_release_metadata(trans
, root
);
785 trans
->block_rsv
= NULL
;
787 if (!list_empty(&trans
->new_bgs
))
788 btrfs_create_pending_block_groups(trans
, root
);
790 if (lock
&& !atomic_read(&root
->fs_info
->open_ioctl_trans
) &&
791 should_end_transaction(trans
, root
) &&
792 ACCESS_ONCE(cur_trans
->state
) == TRANS_STATE_RUNNING
) {
793 spin_lock(&info
->trans_lock
);
794 if (cur_trans
->state
== TRANS_STATE_RUNNING
)
795 cur_trans
->state
= TRANS_STATE_BLOCKED
;
796 spin_unlock(&info
->trans_lock
);
799 if (lock
&& ACCESS_ONCE(cur_trans
->state
) == TRANS_STATE_BLOCKED
) {
801 return btrfs_commit_transaction(trans
, root
);
803 wake_up_process(info
->transaction_kthread
);
806 if (trans
->type
& __TRANS_FREEZABLE
)
807 sb_end_intwrite(root
->fs_info
->sb
);
809 WARN_ON(cur_trans
!= info
->running_transaction
);
810 WARN_ON(atomic_read(&cur_trans
->num_writers
) < 1);
811 atomic_dec(&cur_trans
->num_writers
);
812 extwriter_counter_dec(cur_trans
, trans
->type
);
815 if (waitqueue_active(&cur_trans
->writer_wait
))
816 wake_up(&cur_trans
->writer_wait
);
817 btrfs_put_transaction(cur_trans
);
819 if (current
->journal_info
== trans
)
820 current
->journal_info
= NULL
;
823 btrfs_run_delayed_iputs(root
);
825 if (trans
->aborted
||
826 test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
)) {
827 wake_up_process(info
->transaction_kthread
);
830 assert_qgroups_uptodate(trans
);
832 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
833 if (must_run_delayed_refs
) {
834 btrfs_async_run_delayed_refs(root
, cur
,
835 must_run_delayed_refs
== 1);
840 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
841 struct btrfs_root
*root
)
843 return __btrfs_end_transaction(trans
, root
, 0);
846 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
,
847 struct btrfs_root
*root
)
849 return __btrfs_end_transaction(trans
, root
, 1);
853 * when btree blocks are allocated, they have some corresponding bits set for
854 * them in one of two extent_io trees. This is used to make sure all of
855 * those extents are sent to disk but does not wait on them
857 int btrfs_write_marked_extents(struct btrfs_root
*root
,
858 struct extent_io_tree
*dirty_pages
, int mark
)
862 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
863 struct extent_state
*cached_state
= NULL
;
867 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
868 mark
, &cached_state
)) {
869 bool wait_writeback
= false;
871 err
= convert_extent_bit(dirty_pages
, start
, end
,
873 mark
, &cached_state
, GFP_NOFS
);
875 * convert_extent_bit can return -ENOMEM, which is most of the
876 * time a temporary error. So when it happens, ignore the error
877 * and wait for writeback of this range to finish - because we
878 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
879 * to btrfs_wait_marked_extents() would not know that writeback
880 * for this range started and therefore wouldn't wait for it to
881 * finish - we don't want to commit a superblock that points to
882 * btree nodes/leafs for which writeback hasn't finished yet
883 * (and without errors).
884 * We cleanup any entries left in the io tree when committing
885 * the transaction (through clear_btree_io_tree()).
887 if (err
== -ENOMEM
) {
889 wait_writeback
= true;
892 err
= filemap_fdatawrite_range(mapping
, start
, end
);
895 else if (wait_writeback
)
896 werr
= filemap_fdatawait_range(mapping
, start
, end
);
897 free_extent_state(cached_state
);
906 * when btree blocks are allocated, they have some corresponding bits set for
907 * them in one of two extent_io trees. This is used to make sure all of
908 * those extents are on disk for transaction or log commit. We wait
909 * on all the pages and clear them from the dirty pages state tree
911 int btrfs_wait_marked_extents(struct btrfs_root
*root
,
912 struct extent_io_tree
*dirty_pages
, int mark
)
916 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
917 struct extent_state
*cached_state
= NULL
;
920 struct btrfs_inode
*btree_ino
= BTRFS_I(root
->fs_info
->btree_inode
);
923 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
924 EXTENT_NEED_WAIT
, &cached_state
)) {
926 * Ignore -ENOMEM errors returned by clear_extent_bit().
927 * When committing the transaction, we'll remove any entries
928 * left in the io tree. For a log commit, we don't remove them
929 * after committing the log because the tree can be accessed
930 * concurrently - we do it only at transaction commit time when
931 * it's safe to do it (through clear_btree_io_tree()).
933 err
= clear_extent_bit(dirty_pages
, start
, end
,
935 0, 0, &cached_state
, GFP_NOFS
);
939 err
= filemap_fdatawait_range(mapping
, start
, end
);
942 free_extent_state(cached_state
);
950 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
951 if ((mark
& EXTENT_DIRTY
) &&
952 test_and_clear_bit(BTRFS_INODE_BTREE_LOG1_ERR
,
953 &btree_ino
->runtime_flags
))
956 if ((mark
& EXTENT_NEW
) &&
957 test_and_clear_bit(BTRFS_INODE_BTREE_LOG2_ERR
,
958 &btree_ino
->runtime_flags
))
961 if (test_and_clear_bit(BTRFS_INODE_BTREE_ERR
,
962 &btree_ino
->runtime_flags
))
973 * when btree blocks are allocated, they have some corresponding bits set for
974 * them in one of two extent_io trees. This is used to make sure all of
975 * those extents are on disk for transaction or log commit
977 static int btrfs_write_and_wait_marked_extents(struct btrfs_root
*root
,
978 struct extent_io_tree
*dirty_pages
, int mark
)
982 struct blk_plug plug
;
984 blk_start_plug(&plug
);
985 ret
= btrfs_write_marked_extents(root
, dirty_pages
, mark
);
986 blk_finish_plug(&plug
);
987 ret2
= btrfs_wait_marked_extents(root
, dirty_pages
, mark
);
996 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
,
997 struct btrfs_root
*root
)
1001 ret
= btrfs_write_and_wait_marked_extents(root
,
1002 &trans
->transaction
->dirty_pages
,
1004 clear_btree_io_tree(&trans
->transaction
->dirty_pages
);
1010 * this is used to update the root pointer in the tree of tree roots.
1012 * But, in the case of the extent allocation tree, updating the root
1013 * pointer may allocate blocks which may change the root of the extent
1016 * So, this loops and repeats and makes sure the cowonly root didn't
1017 * change while the root pointer was being updated in the metadata.
1019 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
1020 struct btrfs_root
*root
)
1023 u64 old_root_bytenr
;
1025 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
1026 bool extent_root
= (root
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
);
1028 old_root_used
= btrfs_root_used(&root
->root_item
);
1029 btrfs_write_dirty_block_groups(trans
, root
);
1032 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
1033 if (old_root_bytenr
== root
->node
->start
&&
1034 old_root_used
== btrfs_root_used(&root
->root_item
) &&
1036 list_empty(&trans
->transaction
->dirty_bgs
)))
1039 btrfs_set_root_node(&root
->root_item
, root
->node
);
1040 ret
= btrfs_update_root(trans
, tree_root
,
1046 old_root_used
= btrfs_root_used(&root
->root_item
);
1048 ret
= btrfs_write_dirty_block_groups(trans
, root
);
1052 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1055 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1064 * update all the cowonly tree roots on disk
1066 * The error handling in this function may not be obvious. Any of the
1067 * failures will cause the file system to go offline. We still need
1068 * to clean up the delayed refs.
1070 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
,
1071 struct btrfs_root
*root
)
1073 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1074 struct list_head
*next
;
1075 struct extent_buffer
*eb
;
1078 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
1079 ret
= btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
,
1081 btrfs_tree_unlock(eb
);
1082 free_extent_buffer(eb
);
1087 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1091 ret
= btrfs_run_dev_stats(trans
, root
->fs_info
);
1094 ret
= btrfs_run_dev_replace(trans
, root
->fs_info
);
1097 ret
= btrfs_run_qgroups(trans
, root
->fs_info
);
1101 /* run_qgroups might have added some more refs */
1102 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1106 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
1107 next
= fs_info
->dirty_cowonly_roots
.next
;
1108 list_del_init(next
);
1109 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
1110 clear_bit(BTRFS_ROOT_DIRTY
, &root
->state
);
1112 if (root
!= fs_info
->extent_root
)
1113 list_add_tail(&root
->dirty_list
,
1114 &trans
->transaction
->switch_commits
);
1115 ret
= update_cowonly_root(trans
, root
);
1120 list_add_tail(&fs_info
->extent_root
->dirty_list
,
1121 &trans
->transaction
->switch_commits
);
1122 btrfs_after_dev_replace_commit(fs_info
);
1128 * dead roots are old snapshots that need to be deleted. This allocates
1129 * a dirty root struct and adds it into the list of dead roots that need to
1132 void btrfs_add_dead_root(struct btrfs_root
*root
)
1134 spin_lock(&root
->fs_info
->trans_lock
);
1135 if (list_empty(&root
->root_list
))
1136 list_add_tail(&root
->root_list
, &root
->fs_info
->dead_roots
);
1137 spin_unlock(&root
->fs_info
->trans_lock
);
1141 * update all the cowonly tree roots on disk
1143 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
,
1144 struct btrfs_root
*root
)
1146 struct btrfs_root
*gang
[8];
1147 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1152 spin_lock(&fs_info
->fs_roots_radix_lock
);
1154 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
1157 BTRFS_ROOT_TRANS_TAG
);
1160 for (i
= 0; i
< ret
; i
++) {
1162 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
1163 (unsigned long)root
->root_key
.objectid
,
1164 BTRFS_ROOT_TRANS_TAG
);
1165 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1167 btrfs_free_log(trans
, root
);
1168 btrfs_update_reloc_root(trans
, root
);
1169 btrfs_orphan_commit_root(trans
, root
);
1171 btrfs_save_ino_cache(root
, trans
);
1173 /* see comments in should_cow_block() */
1174 clear_bit(BTRFS_ROOT_FORCE_COW
, &root
->state
);
1175 smp_mb__after_atomic();
1177 if (root
->commit_root
!= root
->node
) {
1178 list_add_tail(&root
->dirty_list
,
1179 &trans
->transaction
->switch_commits
);
1180 btrfs_set_root_node(&root
->root_item
,
1184 err
= btrfs_update_root(trans
, fs_info
->tree_root
,
1187 spin_lock(&fs_info
->fs_roots_radix_lock
);
1192 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1197 * defrag a given btree.
1198 * Every leaf in the btree is read and defragged.
1200 int btrfs_defrag_root(struct btrfs_root
*root
)
1202 struct btrfs_fs_info
*info
= root
->fs_info
;
1203 struct btrfs_trans_handle
*trans
;
1206 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING
, &root
->state
))
1210 trans
= btrfs_start_transaction(root
, 0);
1212 return PTR_ERR(trans
);
1214 ret
= btrfs_defrag_leaves(trans
, root
);
1216 btrfs_end_transaction(trans
, root
);
1217 btrfs_btree_balance_dirty(info
->tree_root
);
1220 if (btrfs_fs_closing(root
->fs_info
) || ret
!= -EAGAIN
)
1223 if (btrfs_defrag_cancelled(root
->fs_info
)) {
1224 pr_debug("BTRFS: defrag_root cancelled\n");
1229 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING
, &root
->state
);
1234 * new snapshots need to be created at a very specific time in the
1235 * transaction commit. This does the actual creation.
1238 * If the error which may affect the commitment of the current transaction
1239 * happens, we should return the error number. If the error which just affect
1240 * the creation of the pending snapshots, just return 0.
1242 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
1243 struct btrfs_fs_info
*fs_info
,
1244 struct btrfs_pending_snapshot
*pending
)
1246 struct btrfs_key key
;
1247 struct btrfs_root_item
*new_root_item
;
1248 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1249 struct btrfs_root
*root
= pending
->root
;
1250 struct btrfs_root
*parent_root
;
1251 struct btrfs_block_rsv
*rsv
;
1252 struct inode
*parent_inode
;
1253 struct btrfs_path
*path
;
1254 struct btrfs_dir_item
*dir_item
;
1255 struct dentry
*dentry
;
1256 struct extent_buffer
*tmp
;
1257 struct extent_buffer
*old
;
1258 struct timespec cur_time
= CURRENT_TIME
;
1266 path
= btrfs_alloc_path();
1268 pending
->error
= -ENOMEM
;
1272 new_root_item
= kmalloc(sizeof(*new_root_item
), GFP_NOFS
);
1273 if (!new_root_item
) {
1274 pending
->error
= -ENOMEM
;
1275 goto root_item_alloc_fail
;
1278 pending
->error
= btrfs_find_free_objectid(tree_root
, &objectid
);
1280 goto no_free_objectid
;
1282 btrfs_reloc_pre_snapshot(trans
, pending
, &to_reserve
);
1284 if (to_reserve
> 0) {
1285 pending
->error
= btrfs_block_rsv_add(root
,
1286 &pending
->block_rsv
,
1288 BTRFS_RESERVE_NO_FLUSH
);
1290 goto no_free_objectid
;
1293 key
.objectid
= objectid
;
1294 key
.offset
= (u64
)-1;
1295 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1297 rsv
= trans
->block_rsv
;
1298 trans
->block_rsv
= &pending
->block_rsv
;
1299 trans
->bytes_reserved
= trans
->block_rsv
->reserved
;
1301 dentry
= pending
->dentry
;
1302 parent_inode
= pending
->dir
;
1303 parent_root
= BTRFS_I(parent_inode
)->root
;
1304 record_root_in_trans(trans
, parent_root
);
1307 * insert the directory item
1309 ret
= btrfs_set_inode_index(parent_inode
, &index
);
1310 BUG_ON(ret
); /* -ENOMEM */
1312 /* check if there is a file/dir which has the same name. */
1313 dir_item
= btrfs_lookup_dir_item(NULL
, parent_root
, path
,
1314 btrfs_ino(parent_inode
),
1315 dentry
->d_name
.name
,
1316 dentry
->d_name
.len
, 0);
1317 if (dir_item
!= NULL
&& !IS_ERR(dir_item
)) {
1318 pending
->error
= -EEXIST
;
1319 goto dir_item_existed
;
1320 } else if (IS_ERR(dir_item
)) {
1321 ret
= PTR_ERR(dir_item
);
1322 btrfs_abort_transaction(trans
, root
, ret
);
1325 btrfs_release_path(path
);
1328 * pull in the delayed directory update
1329 * and the delayed inode item
1330 * otherwise we corrupt the FS during
1333 ret
= btrfs_run_delayed_items(trans
, root
);
1334 if (ret
) { /* Transaction aborted */
1335 btrfs_abort_transaction(trans
, root
, ret
);
1339 record_root_in_trans(trans
, root
);
1340 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
1341 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
1342 btrfs_check_and_init_root_item(new_root_item
);
1344 root_flags
= btrfs_root_flags(new_root_item
);
1345 if (pending
->readonly
)
1346 root_flags
|= BTRFS_ROOT_SUBVOL_RDONLY
;
1348 root_flags
&= ~BTRFS_ROOT_SUBVOL_RDONLY
;
1349 btrfs_set_root_flags(new_root_item
, root_flags
);
1351 btrfs_set_root_generation_v2(new_root_item
,
1353 uuid_le_gen(&new_uuid
);
1354 memcpy(new_root_item
->uuid
, new_uuid
.b
, BTRFS_UUID_SIZE
);
1355 memcpy(new_root_item
->parent_uuid
, root
->root_item
.uuid
,
1357 if (!(root_flags
& BTRFS_ROOT_SUBVOL_RDONLY
)) {
1358 memset(new_root_item
->received_uuid
, 0,
1359 sizeof(new_root_item
->received_uuid
));
1360 memset(&new_root_item
->stime
, 0, sizeof(new_root_item
->stime
));
1361 memset(&new_root_item
->rtime
, 0, sizeof(new_root_item
->rtime
));
1362 btrfs_set_root_stransid(new_root_item
, 0);
1363 btrfs_set_root_rtransid(new_root_item
, 0);
1365 btrfs_set_stack_timespec_sec(&new_root_item
->otime
, cur_time
.tv_sec
);
1366 btrfs_set_stack_timespec_nsec(&new_root_item
->otime
, cur_time
.tv_nsec
);
1367 btrfs_set_root_otransid(new_root_item
, trans
->transid
);
1369 old
= btrfs_lock_root_node(root
);
1370 ret
= btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
);
1372 btrfs_tree_unlock(old
);
1373 free_extent_buffer(old
);
1374 btrfs_abort_transaction(trans
, root
, ret
);
1378 btrfs_set_lock_blocking(old
);
1380 ret
= btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
1381 /* clean up in any case */
1382 btrfs_tree_unlock(old
);
1383 free_extent_buffer(old
);
1385 btrfs_abort_transaction(trans
, root
, ret
);
1390 * We need to flush delayed refs in order to make sure all of our quota
1391 * operations have been done before we call btrfs_qgroup_inherit.
1393 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1395 btrfs_abort_transaction(trans
, root
, ret
);
1399 ret
= btrfs_qgroup_inherit(trans
, fs_info
,
1400 root
->root_key
.objectid
,
1401 objectid
, pending
->inherit
);
1403 btrfs_abort_transaction(trans
, root
, ret
);
1407 /* see comments in should_cow_block() */
1408 set_bit(BTRFS_ROOT_FORCE_COW
, &root
->state
);
1411 btrfs_set_root_node(new_root_item
, tmp
);
1412 /* record when the snapshot was created in key.offset */
1413 key
.offset
= trans
->transid
;
1414 ret
= btrfs_insert_root(trans
, tree_root
, &key
, new_root_item
);
1415 btrfs_tree_unlock(tmp
);
1416 free_extent_buffer(tmp
);
1418 btrfs_abort_transaction(trans
, root
, ret
);
1423 * insert root back/forward references
1425 ret
= btrfs_add_root_ref(trans
, tree_root
, objectid
,
1426 parent_root
->root_key
.objectid
,
1427 btrfs_ino(parent_inode
), index
,
1428 dentry
->d_name
.name
, dentry
->d_name
.len
);
1430 btrfs_abort_transaction(trans
, root
, ret
);
1434 key
.offset
= (u64
)-1;
1435 pending
->snap
= btrfs_read_fs_root_no_name(root
->fs_info
, &key
);
1436 if (IS_ERR(pending
->snap
)) {
1437 ret
= PTR_ERR(pending
->snap
);
1438 btrfs_abort_transaction(trans
, root
, ret
);
1442 ret
= btrfs_reloc_post_snapshot(trans
, pending
);
1444 btrfs_abort_transaction(trans
, root
, ret
);
1448 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1450 btrfs_abort_transaction(trans
, root
, ret
);
1454 ret
= btrfs_insert_dir_item(trans
, parent_root
,
1455 dentry
->d_name
.name
, dentry
->d_name
.len
,
1457 BTRFS_FT_DIR
, index
);
1458 /* We have check then name at the beginning, so it is impossible. */
1459 BUG_ON(ret
== -EEXIST
|| ret
== -EOVERFLOW
);
1461 btrfs_abort_transaction(trans
, root
, ret
);
1465 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
1466 dentry
->d_name
.len
* 2);
1467 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
1468 ret
= btrfs_update_inode_fallback(trans
, parent_root
, parent_inode
);
1470 btrfs_abort_transaction(trans
, root
, ret
);
1473 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
, new_uuid
.b
,
1474 BTRFS_UUID_KEY_SUBVOL
, objectid
);
1476 btrfs_abort_transaction(trans
, root
, ret
);
1479 if (!btrfs_is_empty_uuid(new_root_item
->received_uuid
)) {
1480 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
1481 new_root_item
->received_uuid
,
1482 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
1484 if (ret
&& ret
!= -EEXIST
) {
1485 btrfs_abort_transaction(trans
, root
, ret
);
1490 pending
->error
= ret
;
1492 trans
->block_rsv
= rsv
;
1493 trans
->bytes_reserved
= 0;
1495 kfree(new_root_item
);
1496 root_item_alloc_fail
:
1497 btrfs_free_path(path
);
1502 * create all the snapshots we've scheduled for creation
1504 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
,
1505 struct btrfs_fs_info
*fs_info
)
1507 struct btrfs_pending_snapshot
*pending
, *next
;
1508 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
1511 list_for_each_entry_safe(pending
, next
, head
, list
) {
1512 list_del(&pending
->list
);
1513 ret
= create_pending_snapshot(trans
, fs_info
, pending
);
1520 static void update_super_roots(struct btrfs_root
*root
)
1522 struct btrfs_root_item
*root_item
;
1523 struct btrfs_super_block
*super
;
1525 super
= root
->fs_info
->super_copy
;
1527 root_item
= &root
->fs_info
->chunk_root
->root_item
;
1528 super
->chunk_root
= root_item
->bytenr
;
1529 super
->chunk_root_generation
= root_item
->generation
;
1530 super
->chunk_root_level
= root_item
->level
;
1532 root_item
= &root
->fs_info
->tree_root
->root_item
;
1533 super
->root
= root_item
->bytenr
;
1534 super
->generation
= root_item
->generation
;
1535 super
->root_level
= root_item
->level
;
1536 if (btrfs_test_opt(root
, SPACE_CACHE
))
1537 super
->cache_generation
= root_item
->generation
;
1538 if (root
->fs_info
->update_uuid_tree_gen
)
1539 super
->uuid_tree_generation
= root_item
->generation
;
1542 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
1544 struct btrfs_transaction
*trans
;
1547 spin_lock(&info
->trans_lock
);
1548 trans
= info
->running_transaction
;
1550 ret
= (trans
->state
>= TRANS_STATE_COMMIT_START
);
1551 spin_unlock(&info
->trans_lock
);
1555 int btrfs_transaction_blocked(struct btrfs_fs_info
*info
)
1557 struct btrfs_transaction
*trans
;
1560 spin_lock(&info
->trans_lock
);
1561 trans
= info
->running_transaction
;
1563 ret
= is_transaction_blocked(trans
);
1564 spin_unlock(&info
->trans_lock
);
1569 * wait for the current transaction commit to start and block subsequent
1572 static void wait_current_trans_commit_start(struct btrfs_root
*root
,
1573 struct btrfs_transaction
*trans
)
1575 wait_event(root
->fs_info
->transaction_blocked_wait
,
1576 trans
->state
>= TRANS_STATE_COMMIT_START
||
1581 * wait for the current transaction to start and then become unblocked.
1584 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root
*root
,
1585 struct btrfs_transaction
*trans
)
1587 wait_event(root
->fs_info
->transaction_wait
,
1588 trans
->state
>= TRANS_STATE_UNBLOCKED
||
1593 * commit transactions asynchronously. once btrfs_commit_transaction_async
1594 * returns, any subsequent transaction will not be allowed to join.
1596 struct btrfs_async_commit
{
1597 struct btrfs_trans_handle
*newtrans
;
1598 struct btrfs_root
*root
;
1599 struct work_struct work
;
1602 static void do_async_commit(struct work_struct
*work
)
1604 struct btrfs_async_commit
*ac
=
1605 container_of(work
, struct btrfs_async_commit
, work
);
1608 * We've got freeze protection passed with the transaction.
1609 * Tell lockdep about it.
1611 if (ac
->newtrans
->type
& __TRANS_FREEZABLE
)
1613 &ac
->root
->fs_info
->sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
1616 current
->journal_info
= ac
->newtrans
;
1618 btrfs_commit_transaction(ac
->newtrans
, ac
->root
);
1622 int btrfs_commit_transaction_async(struct btrfs_trans_handle
*trans
,
1623 struct btrfs_root
*root
,
1624 int wait_for_unblock
)
1626 struct btrfs_async_commit
*ac
;
1627 struct btrfs_transaction
*cur_trans
;
1629 ac
= kmalloc(sizeof(*ac
), GFP_NOFS
);
1633 INIT_WORK(&ac
->work
, do_async_commit
);
1635 ac
->newtrans
= btrfs_join_transaction(root
);
1636 if (IS_ERR(ac
->newtrans
)) {
1637 int err
= PTR_ERR(ac
->newtrans
);
1642 /* take transaction reference */
1643 cur_trans
= trans
->transaction
;
1644 atomic_inc(&cur_trans
->use_count
);
1646 btrfs_end_transaction(trans
, root
);
1649 * Tell lockdep we've released the freeze rwsem, since the
1650 * async commit thread will be the one to unlock it.
1652 if (ac
->newtrans
->type
& __TRANS_FREEZABLE
)
1654 &root
->fs_info
->sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
1657 schedule_work(&ac
->work
);
1659 /* wait for transaction to start and unblock */
1660 if (wait_for_unblock
)
1661 wait_current_trans_commit_start_and_unblock(root
, cur_trans
);
1663 wait_current_trans_commit_start(root
, cur_trans
);
1665 if (current
->journal_info
== trans
)
1666 current
->journal_info
= NULL
;
1668 btrfs_put_transaction(cur_trans
);
1673 static void cleanup_transaction(struct btrfs_trans_handle
*trans
,
1674 struct btrfs_root
*root
, int err
)
1676 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1679 WARN_ON(trans
->use_count
> 1);
1681 btrfs_abort_transaction(trans
, root
, err
);
1683 spin_lock(&root
->fs_info
->trans_lock
);
1686 * If the transaction is removed from the list, it means this
1687 * transaction has been committed successfully, so it is impossible
1688 * to call the cleanup function.
1690 BUG_ON(list_empty(&cur_trans
->list
));
1692 list_del_init(&cur_trans
->list
);
1693 if (cur_trans
== root
->fs_info
->running_transaction
) {
1694 cur_trans
->state
= TRANS_STATE_COMMIT_DOING
;
1695 spin_unlock(&root
->fs_info
->trans_lock
);
1696 wait_event(cur_trans
->writer_wait
,
1697 atomic_read(&cur_trans
->num_writers
) == 1);
1699 spin_lock(&root
->fs_info
->trans_lock
);
1701 spin_unlock(&root
->fs_info
->trans_lock
);
1703 btrfs_cleanup_one_transaction(trans
->transaction
, root
);
1705 spin_lock(&root
->fs_info
->trans_lock
);
1706 if (cur_trans
== root
->fs_info
->running_transaction
)
1707 root
->fs_info
->running_transaction
= NULL
;
1708 spin_unlock(&root
->fs_info
->trans_lock
);
1710 if (trans
->type
& __TRANS_FREEZABLE
)
1711 sb_end_intwrite(root
->fs_info
->sb
);
1712 btrfs_put_transaction(cur_trans
);
1713 btrfs_put_transaction(cur_trans
);
1715 trace_btrfs_transaction_commit(root
);
1717 if (current
->journal_info
== trans
)
1718 current
->journal_info
= NULL
;
1719 btrfs_scrub_cancel(root
->fs_info
);
1721 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1724 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info
*fs_info
)
1726 if (btrfs_test_opt(fs_info
->tree_root
, FLUSHONCOMMIT
))
1727 return btrfs_start_delalloc_roots(fs_info
, 1, -1);
1731 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info
*fs_info
)
1733 if (btrfs_test_opt(fs_info
->tree_root
, FLUSHONCOMMIT
))
1734 btrfs_wait_ordered_roots(fs_info
, -1);
1738 btrfs_wait_pending_ordered(struct btrfs_transaction
*cur_trans
,
1739 struct btrfs_fs_info
*fs_info
)
1741 struct btrfs_ordered_extent
*ordered
;
1743 spin_lock(&fs_info
->trans_lock
);
1744 while (!list_empty(&cur_trans
->pending_ordered
)) {
1745 ordered
= list_first_entry(&cur_trans
->pending_ordered
,
1746 struct btrfs_ordered_extent
,
1748 list_del_init(&ordered
->trans_list
);
1749 spin_unlock(&fs_info
->trans_lock
);
1751 wait_event(ordered
->wait
, test_bit(BTRFS_ORDERED_COMPLETE
,
1753 btrfs_put_ordered_extent(ordered
);
1754 spin_lock(&fs_info
->trans_lock
);
1756 spin_unlock(&fs_info
->trans_lock
);
1759 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
,
1760 struct btrfs_root
*root
)
1762 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1763 struct btrfs_transaction
*prev_trans
= NULL
;
1764 struct btrfs_inode
*btree_ino
= BTRFS_I(root
->fs_info
->btree_inode
);
1767 /* Stop the commit early if ->aborted is set */
1768 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1769 ret
= cur_trans
->aborted
;
1770 btrfs_end_transaction(trans
, root
);
1774 /* make a pass through all the delayed refs we have so far
1775 * any runnings procs may add more while we are here
1777 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1779 btrfs_end_transaction(trans
, root
);
1783 btrfs_trans_release_metadata(trans
, root
);
1784 trans
->block_rsv
= NULL
;
1785 if (trans
->qgroup_reserved
) {
1786 btrfs_qgroup_free(root
, trans
->qgroup_reserved
);
1787 trans
->qgroup_reserved
= 0;
1790 cur_trans
= trans
->transaction
;
1793 * set the flushing flag so procs in this transaction have to
1794 * start sending their work down.
1796 cur_trans
->delayed_refs
.flushing
= 1;
1799 if (!list_empty(&trans
->new_bgs
))
1800 btrfs_create_pending_block_groups(trans
, root
);
1802 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1804 btrfs_end_transaction(trans
, root
);
1808 spin_lock(&root
->fs_info
->trans_lock
);
1809 list_splice(&trans
->ordered
, &cur_trans
->pending_ordered
);
1810 if (cur_trans
->state
>= TRANS_STATE_COMMIT_START
) {
1811 spin_unlock(&root
->fs_info
->trans_lock
);
1812 atomic_inc(&cur_trans
->use_count
);
1813 ret
= btrfs_end_transaction(trans
, root
);
1815 wait_for_commit(root
, cur_trans
);
1817 btrfs_put_transaction(cur_trans
);
1822 cur_trans
->state
= TRANS_STATE_COMMIT_START
;
1823 wake_up(&root
->fs_info
->transaction_blocked_wait
);
1825 if (cur_trans
->list
.prev
!= &root
->fs_info
->trans_list
) {
1826 prev_trans
= list_entry(cur_trans
->list
.prev
,
1827 struct btrfs_transaction
, list
);
1828 if (prev_trans
->state
!= TRANS_STATE_COMPLETED
) {
1829 atomic_inc(&prev_trans
->use_count
);
1830 spin_unlock(&root
->fs_info
->trans_lock
);
1832 wait_for_commit(root
, prev_trans
);
1834 btrfs_put_transaction(prev_trans
);
1836 spin_unlock(&root
->fs_info
->trans_lock
);
1839 spin_unlock(&root
->fs_info
->trans_lock
);
1842 extwriter_counter_dec(cur_trans
, trans
->type
);
1844 ret
= btrfs_start_delalloc_flush(root
->fs_info
);
1846 goto cleanup_transaction
;
1848 ret
= btrfs_run_delayed_items(trans
, root
);
1850 goto cleanup_transaction
;
1852 wait_event(cur_trans
->writer_wait
,
1853 extwriter_counter_read(cur_trans
) == 0);
1855 /* some pending stuffs might be added after the previous flush. */
1856 ret
= btrfs_run_delayed_items(trans
, root
);
1858 goto cleanup_transaction
;
1860 btrfs_wait_delalloc_flush(root
->fs_info
);
1862 btrfs_wait_pending_ordered(cur_trans
, root
->fs_info
);
1864 btrfs_scrub_pause(root
);
1866 * Ok now we need to make sure to block out any other joins while we
1867 * commit the transaction. We could have started a join before setting
1868 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
1870 spin_lock(&root
->fs_info
->trans_lock
);
1871 cur_trans
->state
= TRANS_STATE_COMMIT_DOING
;
1872 spin_unlock(&root
->fs_info
->trans_lock
);
1873 wait_event(cur_trans
->writer_wait
,
1874 atomic_read(&cur_trans
->num_writers
) == 1);
1876 /* ->aborted might be set after the previous check, so check it */
1877 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1878 ret
= cur_trans
->aborted
;
1879 goto scrub_continue
;
1882 * the reloc mutex makes sure that we stop
1883 * the balancing code from coming in and moving
1884 * extents around in the middle of the commit
1886 mutex_lock(&root
->fs_info
->reloc_mutex
);
1889 * We needn't worry about the delayed items because we will
1890 * deal with them in create_pending_snapshot(), which is the
1891 * core function of the snapshot creation.
1893 ret
= create_pending_snapshots(trans
, root
->fs_info
);
1895 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1896 goto scrub_continue
;
1900 * We insert the dir indexes of the snapshots and update the inode
1901 * of the snapshots' parents after the snapshot creation, so there
1902 * are some delayed items which are not dealt with. Now deal with
1905 * We needn't worry that this operation will corrupt the snapshots,
1906 * because all the tree which are snapshoted will be forced to COW
1907 * the nodes and leaves.
1909 ret
= btrfs_run_delayed_items(trans
, root
);
1911 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1912 goto scrub_continue
;
1915 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1917 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1918 goto scrub_continue
;
1922 * make sure none of the code above managed to slip in a
1925 btrfs_assert_delayed_root_empty(root
);
1927 WARN_ON(cur_trans
!= trans
->transaction
);
1929 /* btrfs_commit_tree_roots is responsible for getting the
1930 * various roots consistent with each other. Every pointer
1931 * in the tree of tree roots has to point to the most up to date
1932 * root for every subvolume and other tree. So, we have to keep
1933 * the tree logging code from jumping in and changing any
1936 * At this point in the commit, there can't be any tree-log
1937 * writers, but a little lower down we drop the trans mutex
1938 * and let new people in. By holding the tree_log_mutex
1939 * from now until after the super is written, we avoid races
1940 * with the tree-log code.
1942 mutex_lock(&root
->fs_info
->tree_log_mutex
);
1944 ret
= commit_fs_roots(trans
, root
);
1946 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1947 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1948 goto scrub_continue
;
1952 * Since the transaction is done, we can apply the pending changes
1953 * before the next transaction.
1955 btrfs_apply_pending_changes(root
->fs_info
);
1957 /* commit_fs_roots gets rid of all the tree log roots, it is now
1958 * safe to free the root of tree log roots
1960 btrfs_free_log_root_tree(trans
, root
->fs_info
);
1962 ret
= commit_cowonly_roots(trans
, root
);
1964 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1965 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1966 goto scrub_continue
;
1970 * The tasks which save the space cache and inode cache may also
1971 * update ->aborted, check it.
1973 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1974 ret
= cur_trans
->aborted
;
1975 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1976 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1977 goto scrub_continue
;
1980 btrfs_prepare_extent_commit(trans
, root
);
1982 cur_trans
= root
->fs_info
->running_transaction
;
1984 btrfs_set_root_node(&root
->fs_info
->tree_root
->root_item
,
1985 root
->fs_info
->tree_root
->node
);
1986 list_add_tail(&root
->fs_info
->tree_root
->dirty_list
,
1987 &cur_trans
->switch_commits
);
1989 btrfs_set_root_node(&root
->fs_info
->chunk_root
->root_item
,
1990 root
->fs_info
->chunk_root
->node
);
1991 list_add_tail(&root
->fs_info
->chunk_root
->dirty_list
,
1992 &cur_trans
->switch_commits
);
1994 switch_commit_roots(cur_trans
, root
->fs_info
);
1996 assert_qgroups_uptodate(trans
);
1997 ASSERT(list_empty(&cur_trans
->dirty_bgs
));
1998 update_super_roots(root
);
2000 btrfs_set_super_log_root(root
->fs_info
->super_copy
, 0);
2001 btrfs_set_super_log_root_level(root
->fs_info
->super_copy
, 0);
2002 memcpy(root
->fs_info
->super_for_commit
, root
->fs_info
->super_copy
,
2003 sizeof(*root
->fs_info
->super_copy
));
2005 btrfs_update_commit_device_size(root
->fs_info
);
2006 btrfs_update_commit_device_bytes_used(root
, cur_trans
);
2008 clear_bit(BTRFS_INODE_BTREE_LOG1_ERR
, &btree_ino
->runtime_flags
);
2009 clear_bit(BTRFS_INODE_BTREE_LOG2_ERR
, &btree_ino
->runtime_flags
);
2011 spin_lock(&root
->fs_info
->trans_lock
);
2012 cur_trans
->state
= TRANS_STATE_UNBLOCKED
;
2013 root
->fs_info
->running_transaction
= NULL
;
2014 spin_unlock(&root
->fs_info
->trans_lock
);
2015 mutex_unlock(&root
->fs_info
->reloc_mutex
);
2017 wake_up(&root
->fs_info
->transaction_wait
);
2019 ret
= btrfs_write_and_wait_transaction(trans
, root
);
2021 btrfs_error(root
->fs_info
, ret
,
2022 "Error while writing out transaction");
2023 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
2024 goto scrub_continue
;
2027 ret
= write_ctree_super(trans
, root
, 0);
2029 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
2030 goto scrub_continue
;
2034 * the super is written, we can safely allow the tree-loggers
2035 * to go about their business
2037 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
2039 btrfs_finish_extent_commit(trans
, root
);
2041 if (cur_trans
->have_free_bgs
)
2042 btrfs_clear_space_info_full(root
->fs_info
);
2044 root
->fs_info
->last_trans_committed
= cur_trans
->transid
;
2046 * We needn't acquire the lock here because there is no other task
2047 * which can change it.
2049 cur_trans
->state
= TRANS_STATE_COMPLETED
;
2050 wake_up(&cur_trans
->commit_wait
);
2052 spin_lock(&root
->fs_info
->trans_lock
);
2053 list_del_init(&cur_trans
->list
);
2054 spin_unlock(&root
->fs_info
->trans_lock
);
2056 btrfs_put_transaction(cur_trans
);
2057 btrfs_put_transaction(cur_trans
);
2059 if (trans
->type
& __TRANS_FREEZABLE
)
2060 sb_end_intwrite(root
->fs_info
->sb
);
2062 trace_btrfs_transaction_commit(root
);
2064 btrfs_scrub_continue(root
);
2066 if (current
->journal_info
== trans
)
2067 current
->journal_info
= NULL
;
2069 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
2071 if (current
!= root
->fs_info
->transaction_kthread
)
2072 btrfs_run_delayed_iputs(root
);
2077 btrfs_scrub_continue(root
);
2078 cleanup_transaction
:
2079 btrfs_trans_release_metadata(trans
, root
);
2080 trans
->block_rsv
= NULL
;
2081 if (trans
->qgroup_reserved
) {
2082 btrfs_qgroup_free(root
, trans
->qgroup_reserved
);
2083 trans
->qgroup_reserved
= 0;
2085 btrfs_warn(root
->fs_info
, "Skipping commit of aborted transaction.");
2086 if (current
->journal_info
== trans
)
2087 current
->journal_info
= NULL
;
2088 cleanup_transaction(trans
, root
, ret
);
2094 * return < 0 if error
2095 * 0 if there are no more dead_roots at the time of call
2096 * 1 there are more to be processed, call me again
2098 * The return value indicates there are certainly more snapshots to delete, but
2099 * if there comes a new one during processing, it may return 0. We don't mind,
2100 * because btrfs_commit_super will poke cleaner thread and it will process it a
2101 * few seconds later.
2103 int btrfs_clean_one_deleted_snapshot(struct btrfs_root
*root
)
2106 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2108 spin_lock(&fs_info
->trans_lock
);
2109 if (list_empty(&fs_info
->dead_roots
)) {
2110 spin_unlock(&fs_info
->trans_lock
);
2113 root
= list_first_entry(&fs_info
->dead_roots
,
2114 struct btrfs_root
, root_list
);
2115 list_del_init(&root
->root_list
);
2116 spin_unlock(&fs_info
->trans_lock
);
2118 pr_debug("BTRFS: cleaner removing %llu\n", root
->objectid
);
2120 btrfs_kill_all_delayed_nodes(root
);
2122 if (btrfs_header_backref_rev(root
->node
) <
2123 BTRFS_MIXED_BACKREF_REV
)
2124 ret
= btrfs_drop_snapshot(root
, NULL
, 0, 0);
2126 ret
= btrfs_drop_snapshot(root
, NULL
, 1, 0);
2128 return (ret
< 0) ? 0 : 1;
2131 void btrfs_apply_pending_changes(struct btrfs_fs_info
*fs_info
)
2136 prev
= xchg(&fs_info
->pending_changes
, 0);
2140 bit
= 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE
;
2142 btrfs_set_opt(fs_info
->mount_opt
, INODE_MAP_CACHE
);
2145 bit
= 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE
;
2147 btrfs_clear_opt(fs_info
->mount_opt
, INODE_MAP_CACHE
);
2150 bit
= 1 << BTRFS_PENDING_COMMIT
;
2152 btrfs_debug(fs_info
, "pending commit done");
2157 "unknown pending changes left 0x%lx, ignoring", prev
);