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 const 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 if (transaction
->delayed_refs
.pending_csums
)
68 printk(KERN_ERR
"pending csums is %llu\n",
69 transaction
->delayed_refs
.pending_csums
);
70 while (!list_empty(&transaction
->pending_chunks
)) {
71 struct extent_map
*em
;
73 em
= list_first_entry(&transaction
->pending_chunks
,
74 struct extent_map
, list
);
75 list_del_init(&em
->list
);
79 * If any block groups are found in ->deleted_bgs then it's
80 * because the transaction was aborted and a commit did not
81 * happen (things failed before writing the new superblock
82 * and calling btrfs_finish_extent_commit()), so we can not
83 * discard the physical locations of the block groups.
85 while (!list_empty(&transaction
->deleted_bgs
)) {
86 struct btrfs_block_group_cache
*cache
;
88 cache
= list_first_entry(&transaction
->deleted_bgs
,
89 struct btrfs_block_group_cache
,
91 list_del_init(&cache
->bg_list
);
92 btrfs_put_block_group_trimming(cache
);
93 btrfs_put_block_group(cache
);
95 kmem_cache_free(btrfs_transaction_cachep
, transaction
);
99 static void clear_btree_io_tree(struct extent_io_tree
*tree
)
101 spin_lock(&tree
->lock
);
103 * Do a single barrier for the waitqueue_active check here, the state
104 * of the waitqueue should not change once clear_btree_io_tree is
108 while (!RB_EMPTY_ROOT(&tree
->state
)) {
109 struct rb_node
*node
;
110 struct extent_state
*state
;
112 node
= rb_first(&tree
->state
);
113 state
= rb_entry(node
, struct extent_state
, rb_node
);
114 rb_erase(&state
->rb_node
, &tree
->state
);
115 RB_CLEAR_NODE(&state
->rb_node
);
117 * btree io trees aren't supposed to have tasks waiting for
118 * changes in the flags of extent states ever.
120 ASSERT(!waitqueue_active(&state
->wq
));
121 free_extent_state(state
);
123 cond_resched_lock(&tree
->lock
);
125 spin_unlock(&tree
->lock
);
128 static noinline
void switch_commit_roots(struct btrfs_transaction
*trans
,
129 struct btrfs_fs_info
*fs_info
)
131 struct btrfs_root
*root
, *tmp
;
133 down_write(&fs_info
->commit_root_sem
);
134 list_for_each_entry_safe(root
, tmp
, &trans
->switch_commits
,
136 list_del_init(&root
->dirty_list
);
137 free_extent_buffer(root
->commit_root
);
138 root
->commit_root
= btrfs_root_node(root
);
139 if (is_fstree(root
->objectid
))
140 btrfs_unpin_free_ino(root
);
141 clear_btree_io_tree(&root
->dirty_log_pages
);
144 /* We can free old roots now. */
145 spin_lock(&trans
->dropped_roots_lock
);
146 while (!list_empty(&trans
->dropped_roots
)) {
147 root
= list_first_entry(&trans
->dropped_roots
,
148 struct btrfs_root
, root_list
);
149 list_del_init(&root
->root_list
);
150 spin_unlock(&trans
->dropped_roots_lock
);
151 btrfs_drop_and_free_fs_root(fs_info
, root
);
152 spin_lock(&trans
->dropped_roots_lock
);
154 spin_unlock(&trans
->dropped_roots_lock
);
155 up_write(&fs_info
->commit_root_sem
);
158 static inline void extwriter_counter_inc(struct btrfs_transaction
*trans
,
161 if (type
& TRANS_EXTWRITERS
)
162 atomic_inc(&trans
->num_extwriters
);
165 static inline void extwriter_counter_dec(struct btrfs_transaction
*trans
,
168 if (type
& TRANS_EXTWRITERS
)
169 atomic_dec(&trans
->num_extwriters
);
172 static inline void extwriter_counter_init(struct btrfs_transaction
*trans
,
175 atomic_set(&trans
->num_extwriters
, ((type
& TRANS_EXTWRITERS
) ? 1 : 0));
178 static inline int extwriter_counter_read(struct btrfs_transaction
*trans
)
180 return atomic_read(&trans
->num_extwriters
);
184 * either allocate a new transaction or hop into the existing one
186 static noinline
int join_transaction(struct btrfs_root
*root
, unsigned int type
)
188 struct btrfs_transaction
*cur_trans
;
189 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
191 spin_lock(&fs_info
->trans_lock
);
193 /* The file system has been taken offline. No new transactions. */
194 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
195 spin_unlock(&fs_info
->trans_lock
);
199 cur_trans
= fs_info
->running_transaction
;
201 if (cur_trans
->aborted
) {
202 spin_unlock(&fs_info
->trans_lock
);
203 return cur_trans
->aborted
;
205 if (btrfs_blocked_trans_types
[cur_trans
->state
] & type
) {
206 spin_unlock(&fs_info
->trans_lock
);
209 atomic_inc(&cur_trans
->use_count
);
210 atomic_inc(&cur_trans
->num_writers
);
211 extwriter_counter_inc(cur_trans
, type
);
212 spin_unlock(&fs_info
->trans_lock
);
215 spin_unlock(&fs_info
->trans_lock
);
218 * If we are ATTACH, we just want to catch the current transaction,
219 * and commit it. If there is no transaction, just return ENOENT.
221 if (type
== TRANS_ATTACH
)
225 * JOIN_NOLOCK only happens during the transaction commit, so
226 * it is impossible that ->running_transaction is NULL
228 BUG_ON(type
== TRANS_JOIN_NOLOCK
);
230 cur_trans
= kmem_cache_alloc(btrfs_transaction_cachep
, GFP_NOFS
);
234 spin_lock(&fs_info
->trans_lock
);
235 if (fs_info
->running_transaction
) {
237 * someone started a transaction after we unlocked. Make sure
238 * to redo the checks above
240 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
242 } else if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
243 spin_unlock(&fs_info
->trans_lock
);
244 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
248 atomic_set(&cur_trans
->num_writers
, 1);
249 extwriter_counter_init(cur_trans
, type
);
250 init_waitqueue_head(&cur_trans
->writer_wait
);
251 init_waitqueue_head(&cur_trans
->commit_wait
);
252 init_waitqueue_head(&cur_trans
->pending_wait
);
253 cur_trans
->state
= TRANS_STATE_RUNNING
;
255 * One for this trans handle, one so it will live on until we
256 * commit the transaction.
258 atomic_set(&cur_trans
->use_count
, 2);
259 atomic_set(&cur_trans
->pending_ordered
, 0);
260 cur_trans
->flags
= 0;
261 cur_trans
->start_time
= get_seconds();
263 memset(&cur_trans
->delayed_refs
, 0, sizeof(cur_trans
->delayed_refs
));
265 cur_trans
->delayed_refs
.href_root
= RB_ROOT
;
266 cur_trans
->delayed_refs
.dirty_extent_root
= RB_ROOT
;
267 atomic_set(&cur_trans
->delayed_refs
.num_entries
, 0);
270 * although the tree mod log is per file system and not per transaction,
271 * the log must never go across transaction boundaries.
274 if (!list_empty(&fs_info
->tree_mod_seq_list
))
275 WARN(1, KERN_ERR
"BTRFS: tree_mod_seq_list not empty when "
276 "creating a fresh transaction\n");
277 if (!RB_EMPTY_ROOT(&fs_info
->tree_mod_log
))
278 WARN(1, KERN_ERR
"BTRFS: tree_mod_log rb tree not empty when "
279 "creating a fresh transaction\n");
280 atomic64_set(&fs_info
->tree_mod_seq
, 0);
282 spin_lock_init(&cur_trans
->delayed_refs
.lock
);
284 INIT_LIST_HEAD(&cur_trans
->pending_snapshots
);
285 INIT_LIST_HEAD(&cur_trans
->pending_chunks
);
286 INIT_LIST_HEAD(&cur_trans
->switch_commits
);
287 INIT_LIST_HEAD(&cur_trans
->dirty_bgs
);
288 INIT_LIST_HEAD(&cur_trans
->io_bgs
);
289 INIT_LIST_HEAD(&cur_trans
->dropped_roots
);
290 mutex_init(&cur_trans
->cache_write_mutex
);
291 cur_trans
->num_dirty_bgs
= 0;
292 spin_lock_init(&cur_trans
->dirty_bgs_lock
);
293 INIT_LIST_HEAD(&cur_trans
->deleted_bgs
);
294 spin_lock_init(&cur_trans
->dropped_roots_lock
);
295 list_add_tail(&cur_trans
->list
, &fs_info
->trans_list
);
296 extent_io_tree_init(&cur_trans
->dirty_pages
,
297 fs_info
->btree_inode
->i_mapping
);
298 fs_info
->generation
++;
299 cur_trans
->transid
= fs_info
->generation
;
300 fs_info
->running_transaction
= cur_trans
;
301 cur_trans
->aborted
= 0;
302 spin_unlock(&fs_info
->trans_lock
);
308 * this does all the record keeping required to make sure that a reference
309 * counted root is properly recorded in a given transaction. This is required
310 * to make sure the old root from before we joined the transaction is deleted
311 * when the transaction commits
313 static int record_root_in_trans(struct btrfs_trans_handle
*trans
,
314 struct btrfs_root
*root
,
317 if ((test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) &&
318 root
->last_trans
< trans
->transid
) || force
) {
319 WARN_ON(root
== root
->fs_info
->extent_root
);
320 WARN_ON(root
->commit_root
!= root
->node
);
323 * see below for IN_TRANS_SETUP usage rules
324 * we have the reloc mutex held now, so there
325 * is only one writer in this function
327 set_bit(BTRFS_ROOT_IN_TRANS_SETUP
, &root
->state
);
329 /* make sure readers find IN_TRANS_SETUP before
330 * they find our root->last_trans update
334 spin_lock(&root
->fs_info
->fs_roots_radix_lock
);
335 if (root
->last_trans
== trans
->transid
&& !force
) {
336 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
339 radix_tree_tag_set(&root
->fs_info
->fs_roots_radix
,
340 (unsigned long)root
->root_key
.objectid
,
341 BTRFS_ROOT_TRANS_TAG
);
342 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
343 root
->last_trans
= trans
->transid
;
345 /* this is pretty tricky. We don't want to
346 * take the relocation lock in btrfs_record_root_in_trans
347 * unless we're really doing the first setup for this root in
350 * Normally we'd use root->last_trans as a flag to decide
351 * if we want to take the expensive mutex.
353 * But, we have to set root->last_trans before we
354 * init the relocation root, otherwise, we trip over warnings
355 * in ctree.c. The solution used here is to flag ourselves
356 * with root IN_TRANS_SETUP. When this is 1, we're still
357 * fixing up the reloc trees and everyone must wait.
359 * When this is zero, they can trust root->last_trans and fly
360 * through btrfs_record_root_in_trans without having to take the
361 * lock. smp_wmb() makes sure that all the writes above are
362 * done before we pop in the zero below
364 btrfs_init_reloc_root(trans
, root
);
365 smp_mb__before_atomic();
366 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP
, &root
->state
);
372 void btrfs_add_dropped_root(struct btrfs_trans_handle
*trans
,
373 struct btrfs_root
*root
)
375 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
377 /* Add ourselves to the transaction dropped list */
378 spin_lock(&cur_trans
->dropped_roots_lock
);
379 list_add_tail(&root
->root_list
, &cur_trans
->dropped_roots
);
380 spin_unlock(&cur_trans
->dropped_roots_lock
);
382 /* Make sure we don't try to update the root at commit time */
383 spin_lock(&root
->fs_info
->fs_roots_radix_lock
);
384 radix_tree_tag_clear(&root
->fs_info
->fs_roots_radix
,
385 (unsigned long)root
->root_key
.objectid
,
386 BTRFS_ROOT_TRANS_TAG
);
387 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
390 int btrfs_record_root_in_trans(struct btrfs_trans_handle
*trans
,
391 struct btrfs_root
*root
)
393 if (!test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
397 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
401 if (root
->last_trans
== trans
->transid
&&
402 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP
, &root
->state
))
405 mutex_lock(&root
->fs_info
->reloc_mutex
);
406 record_root_in_trans(trans
, root
, 0);
407 mutex_unlock(&root
->fs_info
->reloc_mutex
);
412 static inline int is_transaction_blocked(struct btrfs_transaction
*trans
)
414 return (trans
->state
>= TRANS_STATE_BLOCKED
&&
415 trans
->state
< TRANS_STATE_UNBLOCKED
&&
419 /* wait for commit against the current transaction to become unblocked
420 * when this is done, it is safe to start a new transaction, but the current
421 * transaction might not be fully on disk.
423 static void wait_current_trans(struct btrfs_root
*root
)
425 struct btrfs_transaction
*cur_trans
;
427 spin_lock(&root
->fs_info
->trans_lock
);
428 cur_trans
= root
->fs_info
->running_transaction
;
429 if (cur_trans
&& is_transaction_blocked(cur_trans
)) {
430 atomic_inc(&cur_trans
->use_count
);
431 spin_unlock(&root
->fs_info
->trans_lock
);
433 wait_event(root
->fs_info
->transaction_wait
,
434 cur_trans
->state
>= TRANS_STATE_UNBLOCKED
||
436 btrfs_put_transaction(cur_trans
);
438 spin_unlock(&root
->fs_info
->trans_lock
);
442 static int may_wait_transaction(struct btrfs_root
*root
, int type
)
444 if (root
->fs_info
->log_root_recovering
)
447 if (type
== TRANS_USERSPACE
)
450 if (type
== TRANS_START
&&
451 !atomic_read(&root
->fs_info
->open_ioctl_trans
))
457 static inline bool need_reserve_reloc_root(struct btrfs_root
*root
)
459 if (!root
->fs_info
->reloc_ctl
||
460 !test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
461 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
468 static struct btrfs_trans_handle
*
469 start_transaction(struct btrfs_root
*root
, unsigned int num_items
,
470 unsigned int type
, enum btrfs_reserve_flush_enum flush
)
472 struct btrfs_trans_handle
*h
;
473 struct btrfs_transaction
*cur_trans
;
475 u64 qgroup_reserved
= 0;
476 bool reloc_reserved
= false;
479 /* Send isn't supposed to start transactions. */
480 ASSERT(current
->journal_info
!= BTRFS_SEND_TRANS_STUB
);
482 if (test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
483 return ERR_PTR(-EROFS
);
485 if (current
->journal_info
) {
486 WARN_ON(type
& TRANS_EXTWRITERS
);
487 h
= current
->journal_info
;
489 WARN_ON(h
->use_count
> 2);
490 h
->orig_rsv
= h
->block_rsv
;
496 * Do the reservation before we join the transaction so we can do all
497 * the appropriate flushing if need be.
499 if (num_items
> 0 && root
!= root
->fs_info
->chunk_root
) {
500 qgroup_reserved
= num_items
* root
->nodesize
;
501 ret
= btrfs_qgroup_reserve_meta(root
, qgroup_reserved
);
505 num_bytes
= btrfs_calc_trans_metadata_size(root
, num_items
);
507 * Do the reservation for the relocation root creation
509 if (need_reserve_reloc_root(root
)) {
510 num_bytes
+= root
->nodesize
;
511 reloc_reserved
= true;
514 ret
= btrfs_block_rsv_add(root
,
515 &root
->fs_info
->trans_block_rsv
,
521 h
= kmem_cache_zalloc(btrfs_trans_handle_cachep
, GFP_NOFS
);
528 * If we are JOIN_NOLOCK we're already committing a transaction and
529 * waiting on this guy, so we don't need to do the sb_start_intwrite
530 * because we're already holding a ref. We need this because we could
531 * have raced in and did an fsync() on a file which can kick a commit
532 * and then we deadlock with somebody doing a freeze.
534 * If we are ATTACH, it means we just want to catch the current
535 * transaction and commit it, so we needn't do sb_start_intwrite().
537 if (type
& __TRANS_FREEZABLE
)
538 sb_start_intwrite(root
->fs_info
->sb
);
540 if (may_wait_transaction(root
, type
))
541 wait_current_trans(root
);
544 ret
= join_transaction(root
, type
);
546 wait_current_trans(root
);
547 if (unlikely(type
== TRANS_ATTACH
))
550 } while (ret
== -EBUSY
);
553 /* We must get the transaction if we are JOIN_NOLOCK. */
554 BUG_ON(type
== TRANS_JOIN_NOLOCK
);
558 cur_trans
= root
->fs_info
->running_transaction
;
560 h
->transid
= cur_trans
->transid
;
561 h
->transaction
= cur_trans
;
566 h
->can_flush_pending_bgs
= true;
567 INIT_LIST_HEAD(&h
->qgroup_ref_list
);
568 INIT_LIST_HEAD(&h
->new_bgs
);
571 if (cur_trans
->state
>= TRANS_STATE_BLOCKED
&&
572 may_wait_transaction(root
, type
)) {
573 current
->journal_info
= h
;
574 btrfs_commit_transaction(h
, root
);
579 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
580 h
->transid
, num_bytes
, 1);
581 h
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
582 h
->bytes_reserved
= num_bytes
;
583 h
->reloc_reserved
= reloc_reserved
;
587 btrfs_record_root_in_trans(h
, root
);
589 if (!current
->journal_info
&& type
!= TRANS_USERSPACE
)
590 current
->journal_info
= h
;
594 if (type
& __TRANS_FREEZABLE
)
595 sb_end_intwrite(root
->fs_info
->sb
);
596 kmem_cache_free(btrfs_trans_handle_cachep
, h
);
599 btrfs_block_rsv_release(root
, &root
->fs_info
->trans_block_rsv
,
602 btrfs_qgroup_free_meta(root
, qgroup_reserved
);
606 struct btrfs_trans_handle
*btrfs_start_transaction(struct btrfs_root
*root
,
607 unsigned int num_items
)
609 return start_transaction(root
, num_items
, TRANS_START
,
610 BTRFS_RESERVE_FLUSH_ALL
);
612 struct btrfs_trans_handle
*btrfs_start_transaction_fallback_global_rsv(
613 struct btrfs_root
*root
,
614 unsigned int num_items
,
617 struct btrfs_trans_handle
*trans
;
621 trans
= btrfs_start_transaction(root
, num_items
);
622 if (!IS_ERR(trans
) || PTR_ERR(trans
) != -ENOSPC
)
625 trans
= btrfs_start_transaction(root
, 0);
629 num_bytes
= btrfs_calc_trans_metadata_size(root
, num_items
);
630 ret
= btrfs_cond_migrate_bytes(root
->fs_info
,
631 &root
->fs_info
->trans_block_rsv
,
635 btrfs_end_transaction(trans
, root
);
639 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
640 trans
->bytes_reserved
= num_bytes
;
641 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
642 trans
->transid
, num_bytes
, 1);
647 struct btrfs_trans_handle
*btrfs_start_transaction_lflush(
648 struct btrfs_root
*root
,
649 unsigned int num_items
)
651 return start_transaction(root
, num_items
, TRANS_START
,
652 BTRFS_RESERVE_FLUSH_LIMIT
);
655 struct btrfs_trans_handle
*btrfs_join_transaction(struct btrfs_root
*root
)
657 return start_transaction(root
, 0, TRANS_JOIN
,
658 BTRFS_RESERVE_NO_FLUSH
);
661 struct btrfs_trans_handle
*btrfs_join_transaction_nolock(struct btrfs_root
*root
)
663 return start_transaction(root
, 0, TRANS_JOIN_NOLOCK
,
664 BTRFS_RESERVE_NO_FLUSH
);
667 struct btrfs_trans_handle
*btrfs_start_ioctl_transaction(struct btrfs_root
*root
)
669 return start_transaction(root
, 0, TRANS_USERSPACE
,
670 BTRFS_RESERVE_NO_FLUSH
);
674 * btrfs_attach_transaction() - catch the running transaction
676 * It is used when we want to commit the current the transaction, but
677 * don't want to start a new one.
679 * Note: If this function return -ENOENT, it just means there is no
680 * running transaction. But it is possible that the inactive transaction
681 * is still in the memory, not fully on disk. If you hope there is no
682 * inactive transaction in the fs when -ENOENT is returned, you should
684 * btrfs_attach_transaction_barrier()
686 struct btrfs_trans_handle
*btrfs_attach_transaction(struct btrfs_root
*root
)
688 return start_transaction(root
, 0, TRANS_ATTACH
,
689 BTRFS_RESERVE_NO_FLUSH
);
693 * btrfs_attach_transaction_barrier() - catch the running transaction
695 * It is similar to the above function, the differentia is this one
696 * will wait for all the inactive transactions until they fully
699 struct btrfs_trans_handle
*
700 btrfs_attach_transaction_barrier(struct btrfs_root
*root
)
702 struct btrfs_trans_handle
*trans
;
704 trans
= start_transaction(root
, 0, TRANS_ATTACH
,
705 BTRFS_RESERVE_NO_FLUSH
);
706 if (IS_ERR(trans
) && PTR_ERR(trans
) == -ENOENT
)
707 btrfs_wait_for_commit(root
, 0);
712 /* wait for a transaction commit to be fully complete */
713 static noinline
void wait_for_commit(struct btrfs_root
*root
,
714 struct btrfs_transaction
*commit
)
716 wait_event(commit
->commit_wait
, commit
->state
== TRANS_STATE_COMPLETED
);
719 int btrfs_wait_for_commit(struct btrfs_root
*root
, u64 transid
)
721 struct btrfs_transaction
*cur_trans
= NULL
, *t
;
725 if (transid
<= root
->fs_info
->last_trans_committed
)
728 /* find specified transaction */
729 spin_lock(&root
->fs_info
->trans_lock
);
730 list_for_each_entry(t
, &root
->fs_info
->trans_list
, list
) {
731 if (t
->transid
== transid
) {
733 atomic_inc(&cur_trans
->use_count
);
737 if (t
->transid
> transid
) {
742 spin_unlock(&root
->fs_info
->trans_lock
);
745 * The specified transaction doesn't exist, or we
746 * raced with btrfs_commit_transaction
749 if (transid
> root
->fs_info
->last_trans_committed
)
754 /* find newest transaction that is committing | committed */
755 spin_lock(&root
->fs_info
->trans_lock
);
756 list_for_each_entry_reverse(t
, &root
->fs_info
->trans_list
,
758 if (t
->state
>= TRANS_STATE_COMMIT_START
) {
759 if (t
->state
== TRANS_STATE_COMPLETED
)
762 atomic_inc(&cur_trans
->use_count
);
766 spin_unlock(&root
->fs_info
->trans_lock
);
768 goto out
; /* nothing committing|committed */
771 wait_for_commit(root
, cur_trans
);
772 btrfs_put_transaction(cur_trans
);
777 void btrfs_throttle(struct btrfs_root
*root
)
779 if (!atomic_read(&root
->fs_info
->open_ioctl_trans
))
780 wait_current_trans(root
);
783 static int should_end_transaction(struct btrfs_trans_handle
*trans
,
784 struct btrfs_root
*root
)
786 if (root
->fs_info
->global_block_rsv
.space_info
->full
&&
787 btrfs_check_space_for_delayed_refs(trans
, root
))
790 return !!btrfs_block_rsv_check(root
, &root
->fs_info
->global_block_rsv
, 5);
793 int btrfs_should_end_transaction(struct btrfs_trans_handle
*trans
,
794 struct btrfs_root
*root
)
796 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
801 if (cur_trans
->state
>= TRANS_STATE_BLOCKED
||
802 cur_trans
->delayed_refs
.flushing
)
805 updates
= trans
->delayed_ref_updates
;
806 trans
->delayed_ref_updates
= 0;
808 err
= btrfs_run_delayed_refs(trans
, root
, updates
* 2);
809 if (err
) /* Error code will also eval true */
813 return should_end_transaction(trans
, root
);
816 static int __btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
817 struct btrfs_root
*root
, int throttle
)
819 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
820 struct btrfs_fs_info
*info
= root
->fs_info
;
821 unsigned long cur
= trans
->delayed_ref_updates
;
822 int lock
= (trans
->type
!= TRANS_JOIN_NOLOCK
);
824 int must_run_delayed_refs
= 0;
826 if (trans
->use_count
> 1) {
828 trans
->block_rsv
= trans
->orig_rsv
;
832 btrfs_trans_release_metadata(trans
, root
);
833 trans
->block_rsv
= NULL
;
835 if (!list_empty(&trans
->new_bgs
))
836 btrfs_create_pending_block_groups(trans
, root
);
838 trans
->delayed_ref_updates
= 0;
840 must_run_delayed_refs
=
841 btrfs_should_throttle_delayed_refs(trans
, root
);
842 cur
= max_t(unsigned long, cur
, 32);
845 * don't make the caller wait if they are from a NOLOCK
846 * or ATTACH transaction, it will deadlock with commit
848 if (must_run_delayed_refs
== 1 &&
849 (trans
->type
& (__TRANS_JOIN_NOLOCK
| __TRANS_ATTACH
)))
850 must_run_delayed_refs
= 2;
853 btrfs_trans_release_metadata(trans
, root
);
854 trans
->block_rsv
= NULL
;
856 if (!list_empty(&trans
->new_bgs
))
857 btrfs_create_pending_block_groups(trans
, root
);
859 btrfs_trans_release_chunk_metadata(trans
);
861 if (lock
&& !atomic_read(&root
->fs_info
->open_ioctl_trans
) &&
862 should_end_transaction(trans
, root
) &&
863 ACCESS_ONCE(cur_trans
->state
) == TRANS_STATE_RUNNING
) {
864 spin_lock(&info
->trans_lock
);
865 if (cur_trans
->state
== TRANS_STATE_RUNNING
)
866 cur_trans
->state
= TRANS_STATE_BLOCKED
;
867 spin_unlock(&info
->trans_lock
);
870 if (lock
&& ACCESS_ONCE(cur_trans
->state
) == TRANS_STATE_BLOCKED
) {
872 return btrfs_commit_transaction(trans
, root
);
874 wake_up_process(info
->transaction_kthread
);
877 if (trans
->type
& __TRANS_FREEZABLE
)
878 sb_end_intwrite(root
->fs_info
->sb
);
880 WARN_ON(cur_trans
!= info
->running_transaction
);
881 WARN_ON(atomic_read(&cur_trans
->num_writers
) < 1);
882 atomic_dec(&cur_trans
->num_writers
);
883 extwriter_counter_dec(cur_trans
, trans
->type
);
886 * Make sure counter is updated before we wake up waiters.
889 if (waitqueue_active(&cur_trans
->writer_wait
))
890 wake_up(&cur_trans
->writer_wait
);
891 btrfs_put_transaction(cur_trans
);
893 if (current
->journal_info
== trans
)
894 current
->journal_info
= NULL
;
897 btrfs_run_delayed_iputs(root
);
899 if (trans
->aborted
||
900 test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
)) {
901 wake_up_process(info
->transaction_kthread
);
904 assert_qgroups_uptodate(trans
);
906 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
907 if (must_run_delayed_refs
) {
908 btrfs_async_run_delayed_refs(root
, cur
,
909 must_run_delayed_refs
== 1);
914 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
915 struct btrfs_root
*root
)
917 return __btrfs_end_transaction(trans
, root
, 0);
920 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
,
921 struct btrfs_root
*root
)
923 return __btrfs_end_transaction(trans
, root
, 1);
927 * when btree blocks are allocated, they have some corresponding bits set for
928 * them in one of two extent_io trees. This is used to make sure all of
929 * those extents are sent to disk but does not wait on them
931 int btrfs_write_marked_extents(struct btrfs_root
*root
,
932 struct extent_io_tree
*dirty_pages
, int mark
)
936 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
937 struct extent_state
*cached_state
= NULL
;
941 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
942 mark
, &cached_state
)) {
943 bool wait_writeback
= false;
945 err
= convert_extent_bit(dirty_pages
, start
, end
,
947 mark
, &cached_state
);
949 * convert_extent_bit can return -ENOMEM, which is most of the
950 * time a temporary error. So when it happens, ignore the error
951 * and wait for writeback of this range to finish - because we
952 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
953 * to btrfs_wait_marked_extents() would not know that writeback
954 * for this range started and therefore wouldn't wait for it to
955 * finish - we don't want to commit a superblock that points to
956 * btree nodes/leafs for which writeback hasn't finished yet
957 * (and without errors).
958 * We cleanup any entries left in the io tree when committing
959 * the transaction (through clear_btree_io_tree()).
961 if (err
== -ENOMEM
) {
963 wait_writeback
= true;
966 err
= filemap_fdatawrite_range(mapping
, start
, end
);
969 else if (wait_writeback
)
970 werr
= filemap_fdatawait_range(mapping
, start
, end
);
971 free_extent_state(cached_state
);
980 * when btree blocks are allocated, they have some corresponding bits set for
981 * them in one of two extent_io trees. This is used to make sure all of
982 * those extents are on disk for transaction or log commit. We wait
983 * on all the pages and clear them from the dirty pages state tree
985 int btrfs_wait_marked_extents(struct btrfs_root
*root
,
986 struct extent_io_tree
*dirty_pages
, int mark
)
990 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
991 struct extent_state
*cached_state
= NULL
;
994 struct btrfs_inode
*btree_ino
= BTRFS_I(root
->fs_info
->btree_inode
);
997 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
998 EXTENT_NEED_WAIT
, &cached_state
)) {
1000 * Ignore -ENOMEM errors returned by clear_extent_bit().
1001 * When committing the transaction, we'll remove any entries
1002 * left in the io tree. For a log commit, we don't remove them
1003 * after committing the log because the tree can be accessed
1004 * concurrently - we do it only at transaction commit time when
1005 * it's safe to do it (through clear_btree_io_tree()).
1007 err
= clear_extent_bit(dirty_pages
, start
, end
,
1009 0, 0, &cached_state
, GFP_NOFS
);
1013 err
= filemap_fdatawait_range(mapping
, start
, end
);
1016 free_extent_state(cached_state
);
1017 cached_state
= NULL
;
1024 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
1025 if ((mark
& EXTENT_DIRTY
) &&
1026 test_and_clear_bit(BTRFS_INODE_BTREE_LOG1_ERR
,
1027 &btree_ino
->runtime_flags
))
1030 if ((mark
& EXTENT_NEW
) &&
1031 test_and_clear_bit(BTRFS_INODE_BTREE_LOG2_ERR
,
1032 &btree_ino
->runtime_flags
))
1035 if (test_and_clear_bit(BTRFS_INODE_BTREE_ERR
,
1036 &btree_ino
->runtime_flags
))
1040 if (errors
&& !werr
)
1047 * when btree blocks are allocated, they have some corresponding bits set for
1048 * them in one of two extent_io trees. This is used to make sure all of
1049 * those extents are on disk for transaction or log commit
1051 static int btrfs_write_and_wait_marked_extents(struct btrfs_root
*root
,
1052 struct extent_io_tree
*dirty_pages
, int mark
)
1056 struct blk_plug plug
;
1058 blk_start_plug(&plug
);
1059 ret
= btrfs_write_marked_extents(root
, dirty_pages
, mark
);
1060 blk_finish_plug(&plug
);
1061 ret2
= btrfs_wait_marked_extents(root
, dirty_pages
, mark
);
1070 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
,
1071 struct btrfs_root
*root
)
1075 ret
= btrfs_write_and_wait_marked_extents(root
,
1076 &trans
->transaction
->dirty_pages
,
1078 clear_btree_io_tree(&trans
->transaction
->dirty_pages
);
1084 * this is used to update the root pointer in the tree of tree roots.
1086 * But, in the case of the extent allocation tree, updating the root
1087 * pointer may allocate blocks which may change the root of the extent
1090 * So, this loops and repeats and makes sure the cowonly root didn't
1091 * change while the root pointer was being updated in the metadata.
1093 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
1094 struct btrfs_root
*root
)
1097 u64 old_root_bytenr
;
1099 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
1101 old_root_used
= btrfs_root_used(&root
->root_item
);
1104 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
1105 if (old_root_bytenr
== root
->node
->start
&&
1106 old_root_used
== btrfs_root_used(&root
->root_item
))
1109 btrfs_set_root_node(&root
->root_item
, root
->node
);
1110 ret
= btrfs_update_root(trans
, tree_root
,
1116 old_root_used
= btrfs_root_used(&root
->root_item
);
1123 * update all the cowonly tree roots on disk
1125 * The error handling in this function may not be obvious. Any of the
1126 * failures will cause the file system to go offline. We still need
1127 * to clean up the delayed refs.
1129 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
,
1130 struct btrfs_root
*root
)
1132 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1133 struct list_head
*dirty_bgs
= &trans
->transaction
->dirty_bgs
;
1134 struct list_head
*io_bgs
= &trans
->transaction
->io_bgs
;
1135 struct list_head
*next
;
1136 struct extent_buffer
*eb
;
1139 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
1140 ret
= btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
,
1142 btrfs_tree_unlock(eb
);
1143 free_extent_buffer(eb
);
1148 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1152 ret
= btrfs_run_dev_stats(trans
, root
->fs_info
);
1155 ret
= btrfs_run_dev_replace(trans
, root
->fs_info
);
1158 ret
= btrfs_run_qgroups(trans
, root
->fs_info
);
1162 ret
= btrfs_setup_space_cache(trans
, root
);
1166 /* run_qgroups might have added some more refs */
1167 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1171 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
1172 next
= fs_info
->dirty_cowonly_roots
.next
;
1173 list_del_init(next
);
1174 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
1175 clear_bit(BTRFS_ROOT_DIRTY
, &root
->state
);
1177 if (root
!= fs_info
->extent_root
)
1178 list_add_tail(&root
->dirty_list
,
1179 &trans
->transaction
->switch_commits
);
1180 ret
= update_cowonly_root(trans
, root
);
1183 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1188 while (!list_empty(dirty_bgs
) || !list_empty(io_bgs
)) {
1189 ret
= btrfs_write_dirty_block_groups(trans
, root
);
1192 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1197 if (!list_empty(&fs_info
->dirty_cowonly_roots
))
1200 list_add_tail(&fs_info
->extent_root
->dirty_list
,
1201 &trans
->transaction
->switch_commits
);
1202 btrfs_after_dev_replace_commit(fs_info
);
1208 * dead roots are old snapshots that need to be deleted. This allocates
1209 * a dirty root struct and adds it into the list of dead roots that need to
1212 void btrfs_add_dead_root(struct btrfs_root
*root
)
1214 spin_lock(&root
->fs_info
->trans_lock
);
1215 if (list_empty(&root
->root_list
))
1216 list_add_tail(&root
->root_list
, &root
->fs_info
->dead_roots
);
1217 spin_unlock(&root
->fs_info
->trans_lock
);
1221 * update all the cowonly tree roots on disk
1223 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
,
1224 struct btrfs_root
*root
)
1226 struct btrfs_root
*gang
[8];
1227 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1232 spin_lock(&fs_info
->fs_roots_radix_lock
);
1234 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
1237 BTRFS_ROOT_TRANS_TAG
);
1240 for (i
= 0; i
< ret
; i
++) {
1242 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
1243 (unsigned long)root
->root_key
.objectid
,
1244 BTRFS_ROOT_TRANS_TAG
);
1245 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1247 btrfs_free_log(trans
, root
);
1248 btrfs_update_reloc_root(trans
, root
);
1249 btrfs_orphan_commit_root(trans
, root
);
1251 btrfs_save_ino_cache(root
, trans
);
1253 /* see comments in should_cow_block() */
1254 clear_bit(BTRFS_ROOT_FORCE_COW
, &root
->state
);
1255 smp_mb__after_atomic();
1257 if (root
->commit_root
!= root
->node
) {
1258 list_add_tail(&root
->dirty_list
,
1259 &trans
->transaction
->switch_commits
);
1260 btrfs_set_root_node(&root
->root_item
,
1264 err
= btrfs_update_root(trans
, fs_info
->tree_root
,
1267 spin_lock(&fs_info
->fs_roots_radix_lock
);
1270 btrfs_qgroup_free_meta_all(root
);
1273 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1278 * defrag a given btree.
1279 * Every leaf in the btree is read and defragged.
1281 int btrfs_defrag_root(struct btrfs_root
*root
)
1283 struct btrfs_fs_info
*info
= root
->fs_info
;
1284 struct btrfs_trans_handle
*trans
;
1287 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING
, &root
->state
))
1291 trans
= btrfs_start_transaction(root
, 0);
1293 return PTR_ERR(trans
);
1295 ret
= btrfs_defrag_leaves(trans
, root
);
1297 btrfs_end_transaction(trans
, root
);
1298 btrfs_btree_balance_dirty(info
->tree_root
);
1301 if (btrfs_fs_closing(root
->fs_info
) || ret
!= -EAGAIN
)
1304 if (btrfs_defrag_cancelled(root
->fs_info
)) {
1305 pr_debug("BTRFS: defrag_root cancelled\n");
1310 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING
, &root
->state
);
1315 * Do all special snapshot related qgroup dirty hack.
1317 * Will do all needed qgroup inherit and dirty hack like switch commit
1318 * roots inside one transaction and write all btree into disk, to make
1321 static int qgroup_account_snapshot(struct btrfs_trans_handle
*trans
,
1322 struct btrfs_root
*src
,
1323 struct btrfs_root
*parent
,
1324 struct btrfs_qgroup_inherit
*inherit
,
1327 struct btrfs_fs_info
*fs_info
= src
->fs_info
;
1331 * Save some performance in the case that qgroups are not
1332 * enabled. If this check races with the ioctl, rescan will
1335 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
1336 if (!fs_info
->quota_enabled
) {
1337 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1340 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1343 * We are going to commit transaction, see btrfs_commit_transaction()
1344 * comment for reason locking tree_log_mutex
1346 mutex_lock(&fs_info
->tree_log_mutex
);
1348 ret
= commit_fs_roots(trans
, src
);
1351 ret
= btrfs_qgroup_prepare_account_extents(trans
, fs_info
);
1354 ret
= btrfs_qgroup_account_extents(trans
, fs_info
);
1358 /* Now qgroup are all updated, we can inherit it to new qgroups */
1359 ret
= btrfs_qgroup_inherit(trans
, fs_info
,
1360 src
->root_key
.objectid
, dst_objectid
,
1366 * Now we do a simplified commit transaction, which will:
1367 * 1) commit all subvolume and extent tree
1368 * To ensure all subvolume and extent tree have a valid
1369 * commit_root to accounting later insert_dir_item()
1370 * 2) write all btree blocks onto disk
1371 * This is to make sure later btree modification will be cowed
1372 * Or commit_root can be populated and cause wrong qgroup numbers
1373 * In this simplified commit, we don't really care about other trees
1374 * like chunk and root tree, as they won't affect qgroup.
1375 * And we don't write super to avoid half committed status.
1377 ret
= commit_cowonly_roots(trans
, src
);
1380 switch_commit_roots(trans
->transaction
, fs_info
);
1381 ret
= btrfs_write_and_wait_transaction(trans
, src
);
1383 btrfs_handle_fs_error(fs_info
, ret
,
1384 "Error while writing out transaction for qgroup");
1387 mutex_unlock(&fs_info
->tree_log_mutex
);
1390 * Force parent root to be updated, as we recorded it before so its
1391 * last_trans == cur_transid.
1392 * Or it won't be committed again onto disk after later
1396 record_root_in_trans(trans
, parent
, 1);
1401 * new snapshots need to be created at a very specific time in the
1402 * transaction commit. This does the actual creation.
1405 * If the error which may affect the commitment of the current transaction
1406 * happens, we should return the error number. If the error which just affect
1407 * the creation of the pending snapshots, just return 0.
1409 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
1410 struct btrfs_fs_info
*fs_info
,
1411 struct btrfs_pending_snapshot
*pending
)
1413 struct btrfs_key key
;
1414 struct btrfs_root_item
*new_root_item
;
1415 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1416 struct btrfs_root
*root
= pending
->root
;
1417 struct btrfs_root
*parent_root
;
1418 struct btrfs_block_rsv
*rsv
;
1419 struct inode
*parent_inode
;
1420 struct btrfs_path
*path
;
1421 struct btrfs_dir_item
*dir_item
;
1422 struct dentry
*dentry
;
1423 struct extent_buffer
*tmp
;
1424 struct extent_buffer
*old
;
1425 struct timespec cur_time
;
1433 ASSERT(pending
->path
);
1434 path
= pending
->path
;
1436 ASSERT(pending
->root_item
);
1437 new_root_item
= pending
->root_item
;
1439 pending
->error
= btrfs_find_free_objectid(tree_root
, &objectid
);
1441 goto no_free_objectid
;
1444 * Make qgroup to skip current new snapshot's qgroupid, as it is
1445 * accounted by later btrfs_qgroup_inherit().
1447 btrfs_set_skip_qgroup(trans
, objectid
);
1449 btrfs_reloc_pre_snapshot(pending
, &to_reserve
);
1451 if (to_reserve
> 0) {
1452 pending
->error
= btrfs_block_rsv_add(root
,
1453 &pending
->block_rsv
,
1455 BTRFS_RESERVE_NO_FLUSH
);
1457 goto clear_skip_qgroup
;
1460 key
.objectid
= objectid
;
1461 key
.offset
= (u64
)-1;
1462 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1464 rsv
= trans
->block_rsv
;
1465 trans
->block_rsv
= &pending
->block_rsv
;
1466 trans
->bytes_reserved
= trans
->block_rsv
->reserved
;
1467 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
1469 trans
->bytes_reserved
, 1);
1470 dentry
= pending
->dentry
;
1471 parent_inode
= pending
->dir
;
1472 parent_root
= BTRFS_I(parent_inode
)->root
;
1473 record_root_in_trans(trans
, parent_root
, 0);
1475 cur_time
= current_fs_time(parent_inode
->i_sb
);
1478 * insert the directory item
1480 ret
= btrfs_set_inode_index(parent_inode
, &index
);
1481 BUG_ON(ret
); /* -ENOMEM */
1483 /* check if there is a file/dir which has the same name. */
1484 dir_item
= btrfs_lookup_dir_item(NULL
, parent_root
, path
,
1485 btrfs_ino(parent_inode
),
1486 dentry
->d_name
.name
,
1487 dentry
->d_name
.len
, 0);
1488 if (dir_item
!= NULL
&& !IS_ERR(dir_item
)) {
1489 pending
->error
= -EEXIST
;
1490 goto dir_item_existed
;
1491 } else if (IS_ERR(dir_item
)) {
1492 ret
= PTR_ERR(dir_item
);
1493 btrfs_abort_transaction(trans
, root
, ret
);
1496 btrfs_release_path(path
);
1499 * pull in the delayed directory update
1500 * and the delayed inode item
1501 * otherwise we corrupt the FS during
1504 ret
= btrfs_run_delayed_items(trans
, root
);
1505 if (ret
) { /* Transaction aborted */
1506 btrfs_abort_transaction(trans
, root
, ret
);
1510 record_root_in_trans(trans
, root
, 0);
1511 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
1512 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
1513 btrfs_check_and_init_root_item(new_root_item
);
1515 root_flags
= btrfs_root_flags(new_root_item
);
1516 if (pending
->readonly
)
1517 root_flags
|= BTRFS_ROOT_SUBVOL_RDONLY
;
1519 root_flags
&= ~BTRFS_ROOT_SUBVOL_RDONLY
;
1520 btrfs_set_root_flags(new_root_item
, root_flags
);
1522 btrfs_set_root_generation_v2(new_root_item
,
1524 uuid_le_gen(&new_uuid
);
1525 memcpy(new_root_item
->uuid
, new_uuid
.b
, BTRFS_UUID_SIZE
);
1526 memcpy(new_root_item
->parent_uuid
, root
->root_item
.uuid
,
1528 if (!(root_flags
& BTRFS_ROOT_SUBVOL_RDONLY
)) {
1529 memset(new_root_item
->received_uuid
, 0,
1530 sizeof(new_root_item
->received_uuid
));
1531 memset(&new_root_item
->stime
, 0, sizeof(new_root_item
->stime
));
1532 memset(&new_root_item
->rtime
, 0, sizeof(new_root_item
->rtime
));
1533 btrfs_set_root_stransid(new_root_item
, 0);
1534 btrfs_set_root_rtransid(new_root_item
, 0);
1536 btrfs_set_stack_timespec_sec(&new_root_item
->otime
, cur_time
.tv_sec
);
1537 btrfs_set_stack_timespec_nsec(&new_root_item
->otime
, cur_time
.tv_nsec
);
1538 btrfs_set_root_otransid(new_root_item
, trans
->transid
);
1540 old
= btrfs_lock_root_node(root
);
1541 ret
= btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
);
1543 btrfs_tree_unlock(old
);
1544 free_extent_buffer(old
);
1545 btrfs_abort_transaction(trans
, root
, ret
);
1549 btrfs_set_lock_blocking(old
);
1551 ret
= btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
1552 /* clean up in any case */
1553 btrfs_tree_unlock(old
);
1554 free_extent_buffer(old
);
1556 btrfs_abort_transaction(trans
, root
, ret
);
1559 /* see comments in should_cow_block() */
1560 set_bit(BTRFS_ROOT_FORCE_COW
, &root
->state
);
1563 btrfs_set_root_node(new_root_item
, tmp
);
1564 /* record when the snapshot was created in key.offset */
1565 key
.offset
= trans
->transid
;
1566 ret
= btrfs_insert_root(trans
, tree_root
, &key
, new_root_item
);
1567 btrfs_tree_unlock(tmp
);
1568 free_extent_buffer(tmp
);
1570 btrfs_abort_transaction(trans
, root
, ret
);
1575 * insert root back/forward references
1577 ret
= btrfs_add_root_ref(trans
, tree_root
, objectid
,
1578 parent_root
->root_key
.objectid
,
1579 btrfs_ino(parent_inode
), index
,
1580 dentry
->d_name
.name
, dentry
->d_name
.len
);
1582 btrfs_abort_transaction(trans
, root
, ret
);
1586 key
.offset
= (u64
)-1;
1587 pending
->snap
= btrfs_read_fs_root_no_name(root
->fs_info
, &key
);
1588 if (IS_ERR(pending
->snap
)) {
1589 ret
= PTR_ERR(pending
->snap
);
1590 btrfs_abort_transaction(trans
, root
, ret
);
1594 ret
= btrfs_reloc_post_snapshot(trans
, pending
);
1596 btrfs_abort_transaction(trans
, root
, ret
);
1600 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1602 btrfs_abort_transaction(trans
, root
, ret
);
1607 * Do special qgroup accounting for snapshot, as we do some qgroup
1608 * snapshot hack to do fast snapshot.
1609 * To co-operate with that hack, we do hack again.
1610 * Or snapshot will be greatly slowed down by a subtree qgroup rescan
1612 ret
= qgroup_account_snapshot(trans
, root
, parent_root
,
1613 pending
->inherit
, objectid
);
1617 ret
= btrfs_insert_dir_item(trans
, parent_root
,
1618 dentry
->d_name
.name
, dentry
->d_name
.len
,
1620 BTRFS_FT_DIR
, index
);
1621 /* We have check then name at the beginning, so it is impossible. */
1622 BUG_ON(ret
== -EEXIST
|| ret
== -EOVERFLOW
);
1624 btrfs_abort_transaction(trans
, root
, ret
);
1628 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
1629 dentry
->d_name
.len
* 2);
1630 parent_inode
->i_mtime
= parent_inode
->i_ctime
=
1631 current_fs_time(parent_inode
->i_sb
);
1632 ret
= btrfs_update_inode_fallback(trans
, parent_root
, parent_inode
);
1634 btrfs_abort_transaction(trans
, root
, ret
);
1637 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
, new_uuid
.b
,
1638 BTRFS_UUID_KEY_SUBVOL
, objectid
);
1640 btrfs_abort_transaction(trans
, root
, ret
);
1643 if (!btrfs_is_empty_uuid(new_root_item
->received_uuid
)) {
1644 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
1645 new_root_item
->received_uuid
,
1646 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
1648 if (ret
&& ret
!= -EEXIST
) {
1649 btrfs_abort_transaction(trans
, root
, ret
);
1654 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1656 btrfs_abort_transaction(trans
, root
, ret
);
1661 pending
->error
= ret
;
1663 trans
->block_rsv
= rsv
;
1664 trans
->bytes_reserved
= 0;
1666 btrfs_clear_skip_qgroup(trans
);
1668 kfree(new_root_item
);
1669 pending
->root_item
= NULL
;
1670 btrfs_free_path(path
);
1671 pending
->path
= NULL
;
1677 * create all the snapshots we've scheduled for creation
1679 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
,
1680 struct btrfs_fs_info
*fs_info
)
1682 struct btrfs_pending_snapshot
*pending
, *next
;
1683 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
1686 list_for_each_entry_safe(pending
, next
, head
, list
) {
1687 list_del(&pending
->list
);
1688 ret
= create_pending_snapshot(trans
, fs_info
, pending
);
1695 static void update_super_roots(struct btrfs_root
*root
)
1697 struct btrfs_root_item
*root_item
;
1698 struct btrfs_super_block
*super
;
1700 super
= root
->fs_info
->super_copy
;
1702 root_item
= &root
->fs_info
->chunk_root
->root_item
;
1703 super
->chunk_root
= root_item
->bytenr
;
1704 super
->chunk_root_generation
= root_item
->generation
;
1705 super
->chunk_root_level
= root_item
->level
;
1707 root_item
= &root
->fs_info
->tree_root
->root_item
;
1708 super
->root
= root_item
->bytenr
;
1709 super
->generation
= root_item
->generation
;
1710 super
->root_level
= root_item
->level
;
1711 if (btrfs_test_opt(root
, SPACE_CACHE
))
1712 super
->cache_generation
= root_item
->generation
;
1713 if (root
->fs_info
->update_uuid_tree_gen
)
1714 super
->uuid_tree_generation
= root_item
->generation
;
1717 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
1719 struct btrfs_transaction
*trans
;
1722 spin_lock(&info
->trans_lock
);
1723 trans
= info
->running_transaction
;
1725 ret
= (trans
->state
>= TRANS_STATE_COMMIT_START
);
1726 spin_unlock(&info
->trans_lock
);
1730 int btrfs_transaction_blocked(struct btrfs_fs_info
*info
)
1732 struct btrfs_transaction
*trans
;
1735 spin_lock(&info
->trans_lock
);
1736 trans
= info
->running_transaction
;
1738 ret
= is_transaction_blocked(trans
);
1739 spin_unlock(&info
->trans_lock
);
1744 * wait for the current transaction commit to start and block subsequent
1747 static void wait_current_trans_commit_start(struct btrfs_root
*root
,
1748 struct btrfs_transaction
*trans
)
1750 wait_event(root
->fs_info
->transaction_blocked_wait
,
1751 trans
->state
>= TRANS_STATE_COMMIT_START
||
1756 * wait for the current transaction to start and then become unblocked.
1759 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root
*root
,
1760 struct btrfs_transaction
*trans
)
1762 wait_event(root
->fs_info
->transaction_wait
,
1763 trans
->state
>= TRANS_STATE_UNBLOCKED
||
1768 * commit transactions asynchronously. once btrfs_commit_transaction_async
1769 * returns, any subsequent transaction will not be allowed to join.
1771 struct btrfs_async_commit
{
1772 struct btrfs_trans_handle
*newtrans
;
1773 struct btrfs_root
*root
;
1774 struct work_struct work
;
1777 static void do_async_commit(struct work_struct
*work
)
1779 struct btrfs_async_commit
*ac
=
1780 container_of(work
, struct btrfs_async_commit
, work
);
1783 * We've got freeze protection passed with the transaction.
1784 * Tell lockdep about it.
1786 if (ac
->newtrans
->type
& __TRANS_FREEZABLE
)
1787 __sb_writers_acquired(ac
->root
->fs_info
->sb
, SB_FREEZE_FS
);
1789 current
->journal_info
= ac
->newtrans
;
1791 btrfs_commit_transaction(ac
->newtrans
, ac
->root
);
1795 int btrfs_commit_transaction_async(struct btrfs_trans_handle
*trans
,
1796 struct btrfs_root
*root
,
1797 int wait_for_unblock
)
1799 struct btrfs_async_commit
*ac
;
1800 struct btrfs_transaction
*cur_trans
;
1802 ac
= kmalloc(sizeof(*ac
), GFP_NOFS
);
1806 INIT_WORK(&ac
->work
, do_async_commit
);
1808 ac
->newtrans
= btrfs_join_transaction(root
);
1809 if (IS_ERR(ac
->newtrans
)) {
1810 int err
= PTR_ERR(ac
->newtrans
);
1815 /* take transaction reference */
1816 cur_trans
= trans
->transaction
;
1817 atomic_inc(&cur_trans
->use_count
);
1819 btrfs_end_transaction(trans
, root
);
1822 * Tell lockdep we've released the freeze rwsem, since the
1823 * async commit thread will be the one to unlock it.
1825 if (ac
->newtrans
->type
& __TRANS_FREEZABLE
)
1826 __sb_writers_release(root
->fs_info
->sb
, SB_FREEZE_FS
);
1828 schedule_work(&ac
->work
);
1830 /* wait for transaction to start and unblock */
1831 if (wait_for_unblock
)
1832 wait_current_trans_commit_start_and_unblock(root
, cur_trans
);
1834 wait_current_trans_commit_start(root
, cur_trans
);
1836 if (current
->journal_info
== trans
)
1837 current
->journal_info
= NULL
;
1839 btrfs_put_transaction(cur_trans
);
1844 static void cleanup_transaction(struct btrfs_trans_handle
*trans
,
1845 struct btrfs_root
*root
, int err
)
1847 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1850 WARN_ON(trans
->use_count
> 1);
1852 btrfs_abort_transaction(trans
, root
, err
);
1854 spin_lock(&root
->fs_info
->trans_lock
);
1857 * If the transaction is removed from the list, it means this
1858 * transaction has been committed successfully, so it is impossible
1859 * to call the cleanup function.
1861 BUG_ON(list_empty(&cur_trans
->list
));
1863 list_del_init(&cur_trans
->list
);
1864 if (cur_trans
== root
->fs_info
->running_transaction
) {
1865 cur_trans
->state
= TRANS_STATE_COMMIT_DOING
;
1866 spin_unlock(&root
->fs_info
->trans_lock
);
1867 wait_event(cur_trans
->writer_wait
,
1868 atomic_read(&cur_trans
->num_writers
) == 1);
1870 spin_lock(&root
->fs_info
->trans_lock
);
1872 spin_unlock(&root
->fs_info
->trans_lock
);
1874 btrfs_cleanup_one_transaction(trans
->transaction
, root
);
1876 spin_lock(&root
->fs_info
->trans_lock
);
1877 if (cur_trans
== root
->fs_info
->running_transaction
)
1878 root
->fs_info
->running_transaction
= NULL
;
1879 spin_unlock(&root
->fs_info
->trans_lock
);
1881 if (trans
->type
& __TRANS_FREEZABLE
)
1882 sb_end_intwrite(root
->fs_info
->sb
);
1883 btrfs_put_transaction(cur_trans
);
1884 btrfs_put_transaction(cur_trans
);
1886 trace_btrfs_transaction_commit(root
);
1888 if (current
->journal_info
== trans
)
1889 current
->journal_info
= NULL
;
1890 btrfs_scrub_cancel(root
->fs_info
);
1892 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1895 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info
*fs_info
)
1897 if (btrfs_test_opt(fs_info
->tree_root
, FLUSHONCOMMIT
))
1898 return btrfs_start_delalloc_roots(fs_info
, 1, -1);
1902 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info
*fs_info
)
1904 if (btrfs_test_opt(fs_info
->tree_root
, FLUSHONCOMMIT
))
1905 btrfs_wait_ordered_roots(fs_info
, -1, 0, (u64
)-1);
1909 btrfs_wait_pending_ordered(struct btrfs_transaction
*cur_trans
)
1911 wait_event(cur_trans
->pending_wait
,
1912 atomic_read(&cur_trans
->pending_ordered
) == 0);
1915 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
,
1916 struct btrfs_root
*root
)
1918 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1919 struct btrfs_transaction
*prev_trans
= NULL
;
1920 struct btrfs_inode
*btree_ino
= BTRFS_I(root
->fs_info
->btree_inode
);
1923 /* Stop the commit early if ->aborted is set */
1924 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1925 ret
= cur_trans
->aborted
;
1926 btrfs_end_transaction(trans
, root
);
1930 /* make a pass through all the delayed refs we have so far
1931 * any runnings procs may add more while we are here
1933 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1935 btrfs_end_transaction(trans
, root
);
1939 btrfs_trans_release_metadata(trans
, root
);
1940 trans
->block_rsv
= NULL
;
1942 cur_trans
= trans
->transaction
;
1945 * set the flushing flag so procs in this transaction have to
1946 * start sending their work down.
1948 cur_trans
->delayed_refs
.flushing
= 1;
1951 if (!list_empty(&trans
->new_bgs
))
1952 btrfs_create_pending_block_groups(trans
, root
);
1954 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1956 btrfs_end_transaction(trans
, root
);
1960 if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN
, &cur_trans
->flags
)) {
1963 /* this mutex is also taken before trying to set
1964 * block groups readonly. We need to make sure
1965 * that nobody has set a block group readonly
1966 * after a extents from that block group have been
1967 * allocated for cache files. btrfs_set_block_group_ro
1968 * will wait for the transaction to commit if it
1969 * finds BTRFS_TRANS_DIRTY_BG_RUN set.
1971 * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure
1972 * only one process starts all the block group IO. It wouldn't
1973 * hurt to have more than one go through, but there's no
1974 * real advantage to it either.
1976 mutex_lock(&root
->fs_info
->ro_block_group_mutex
);
1977 if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN
,
1980 mutex_unlock(&root
->fs_info
->ro_block_group_mutex
);
1983 ret
= btrfs_start_dirty_block_groups(trans
, root
);
1986 btrfs_end_transaction(trans
, root
);
1990 spin_lock(&root
->fs_info
->trans_lock
);
1991 if (cur_trans
->state
>= TRANS_STATE_COMMIT_START
) {
1992 spin_unlock(&root
->fs_info
->trans_lock
);
1993 atomic_inc(&cur_trans
->use_count
);
1994 ret
= btrfs_end_transaction(trans
, root
);
1996 wait_for_commit(root
, cur_trans
);
1998 if (unlikely(cur_trans
->aborted
))
1999 ret
= cur_trans
->aborted
;
2001 btrfs_put_transaction(cur_trans
);
2006 cur_trans
->state
= TRANS_STATE_COMMIT_START
;
2007 wake_up(&root
->fs_info
->transaction_blocked_wait
);
2009 if (cur_trans
->list
.prev
!= &root
->fs_info
->trans_list
) {
2010 prev_trans
= list_entry(cur_trans
->list
.prev
,
2011 struct btrfs_transaction
, list
);
2012 if (prev_trans
->state
!= TRANS_STATE_COMPLETED
) {
2013 atomic_inc(&prev_trans
->use_count
);
2014 spin_unlock(&root
->fs_info
->trans_lock
);
2016 wait_for_commit(root
, prev_trans
);
2017 ret
= prev_trans
->aborted
;
2019 btrfs_put_transaction(prev_trans
);
2021 goto cleanup_transaction
;
2023 spin_unlock(&root
->fs_info
->trans_lock
);
2026 spin_unlock(&root
->fs_info
->trans_lock
);
2029 extwriter_counter_dec(cur_trans
, trans
->type
);
2031 ret
= btrfs_start_delalloc_flush(root
->fs_info
);
2033 goto cleanup_transaction
;
2035 ret
= btrfs_run_delayed_items(trans
, root
);
2037 goto cleanup_transaction
;
2039 wait_event(cur_trans
->writer_wait
,
2040 extwriter_counter_read(cur_trans
) == 0);
2042 /* some pending stuffs might be added after the previous flush. */
2043 ret
= btrfs_run_delayed_items(trans
, root
);
2045 goto cleanup_transaction
;
2047 btrfs_wait_delalloc_flush(root
->fs_info
);
2049 btrfs_wait_pending_ordered(cur_trans
);
2051 btrfs_scrub_pause(root
);
2053 * Ok now we need to make sure to block out any other joins while we
2054 * commit the transaction. We could have started a join before setting
2055 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
2057 spin_lock(&root
->fs_info
->trans_lock
);
2058 cur_trans
->state
= TRANS_STATE_COMMIT_DOING
;
2059 spin_unlock(&root
->fs_info
->trans_lock
);
2060 wait_event(cur_trans
->writer_wait
,
2061 atomic_read(&cur_trans
->num_writers
) == 1);
2063 /* ->aborted might be set after the previous check, so check it */
2064 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
2065 ret
= cur_trans
->aborted
;
2066 goto scrub_continue
;
2069 * the reloc mutex makes sure that we stop
2070 * the balancing code from coming in and moving
2071 * extents around in the middle of the commit
2073 mutex_lock(&root
->fs_info
->reloc_mutex
);
2076 * We needn't worry about the delayed items because we will
2077 * deal with them in create_pending_snapshot(), which is the
2078 * core function of the snapshot creation.
2080 ret
= create_pending_snapshots(trans
, root
->fs_info
);
2082 mutex_unlock(&root
->fs_info
->reloc_mutex
);
2083 goto scrub_continue
;
2087 * We insert the dir indexes of the snapshots and update the inode
2088 * of the snapshots' parents after the snapshot creation, so there
2089 * are some delayed items which are not dealt with. Now deal with
2092 * We needn't worry that this operation will corrupt the snapshots,
2093 * because all the tree which are snapshoted will be forced to COW
2094 * the nodes and leaves.
2096 ret
= btrfs_run_delayed_items(trans
, root
);
2098 mutex_unlock(&root
->fs_info
->reloc_mutex
);
2099 goto scrub_continue
;
2102 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
2104 mutex_unlock(&root
->fs_info
->reloc_mutex
);
2105 goto scrub_continue
;
2108 /* Reocrd old roots for later qgroup accounting */
2109 ret
= btrfs_qgroup_prepare_account_extents(trans
, root
->fs_info
);
2111 mutex_unlock(&root
->fs_info
->reloc_mutex
);
2112 goto scrub_continue
;
2116 * make sure none of the code above managed to slip in a
2119 btrfs_assert_delayed_root_empty(root
);
2121 WARN_ON(cur_trans
!= trans
->transaction
);
2123 /* btrfs_commit_tree_roots is responsible for getting the
2124 * various roots consistent with each other. Every pointer
2125 * in the tree of tree roots has to point to the most up to date
2126 * root for every subvolume and other tree. So, we have to keep
2127 * the tree logging code from jumping in and changing any
2130 * At this point in the commit, there can't be any tree-log
2131 * writers, but a little lower down we drop the trans mutex
2132 * and let new people in. By holding the tree_log_mutex
2133 * from now until after the super is written, we avoid races
2134 * with the tree-log code.
2136 mutex_lock(&root
->fs_info
->tree_log_mutex
);
2138 ret
= commit_fs_roots(trans
, root
);
2140 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
2141 mutex_unlock(&root
->fs_info
->reloc_mutex
);
2142 goto scrub_continue
;
2146 * Since the transaction is done, we can apply the pending changes
2147 * before the next transaction.
2149 btrfs_apply_pending_changes(root
->fs_info
);
2151 /* commit_fs_roots gets rid of all the tree log roots, it is now
2152 * safe to free the root of tree log roots
2154 btrfs_free_log_root_tree(trans
, root
->fs_info
);
2157 * Since fs roots are all committed, we can get a quite accurate
2158 * new_roots. So let's do quota accounting.
2160 ret
= btrfs_qgroup_account_extents(trans
, root
->fs_info
);
2162 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
2163 mutex_unlock(&root
->fs_info
->reloc_mutex
);
2164 goto scrub_continue
;
2167 ret
= commit_cowonly_roots(trans
, root
);
2169 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
2170 mutex_unlock(&root
->fs_info
->reloc_mutex
);
2171 goto scrub_continue
;
2175 * The tasks which save the space cache and inode cache may also
2176 * update ->aborted, check it.
2178 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
2179 ret
= cur_trans
->aborted
;
2180 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
2181 mutex_unlock(&root
->fs_info
->reloc_mutex
);
2182 goto scrub_continue
;
2185 btrfs_prepare_extent_commit(trans
, root
);
2187 cur_trans
= root
->fs_info
->running_transaction
;
2189 btrfs_set_root_node(&root
->fs_info
->tree_root
->root_item
,
2190 root
->fs_info
->tree_root
->node
);
2191 list_add_tail(&root
->fs_info
->tree_root
->dirty_list
,
2192 &cur_trans
->switch_commits
);
2194 btrfs_set_root_node(&root
->fs_info
->chunk_root
->root_item
,
2195 root
->fs_info
->chunk_root
->node
);
2196 list_add_tail(&root
->fs_info
->chunk_root
->dirty_list
,
2197 &cur_trans
->switch_commits
);
2199 switch_commit_roots(cur_trans
, root
->fs_info
);
2201 assert_qgroups_uptodate(trans
);
2202 ASSERT(list_empty(&cur_trans
->dirty_bgs
));
2203 ASSERT(list_empty(&cur_trans
->io_bgs
));
2204 update_super_roots(root
);
2206 btrfs_set_super_log_root(root
->fs_info
->super_copy
, 0);
2207 btrfs_set_super_log_root_level(root
->fs_info
->super_copy
, 0);
2208 memcpy(root
->fs_info
->super_for_commit
, root
->fs_info
->super_copy
,
2209 sizeof(*root
->fs_info
->super_copy
));
2211 btrfs_update_commit_device_size(root
->fs_info
);
2212 btrfs_update_commit_device_bytes_used(root
, cur_trans
);
2214 clear_bit(BTRFS_INODE_BTREE_LOG1_ERR
, &btree_ino
->runtime_flags
);
2215 clear_bit(BTRFS_INODE_BTREE_LOG2_ERR
, &btree_ino
->runtime_flags
);
2217 btrfs_trans_release_chunk_metadata(trans
);
2219 spin_lock(&root
->fs_info
->trans_lock
);
2220 cur_trans
->state
= TRANS_STATE_UNBLOCKED
;
2221 root
->fs_info
->running_transaction
= NULL
;
2222 spin_unlock(&root
->fs_info
->trans_lock
);
2223 mutex_unlock(&root
->fs_info
->reloc_mutex
);
2225 wake_up(&root
->fs_info
->transaction_wait
);
2227 ret
= btrfs_write_and_wait_transaction(trans
, root
);
2229 btrfs_handle_fs_error(root
->fs_info
, ret
,
2230 "Error while writing out transaction");
2231 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
2232 goto scrub_continue
;
2235 ret
= write_ctree_super(trans
, root
, 0);
2237 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
2238 goto scrub_continue
;
2242 * the super is written, we can safely allow the tree-loggers
2243 * to go about their business
2245 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
2247 btrfs_finish_extent_commit(trans
, root
);
2249 if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS
, &cur_trans
->flags
))
2250 btrfs_clear_space_info_full(root
->fs_info
);
2252 root
->fs_info
->last_trans_committed
= cur_trans
->transid
;
2254 * We needn't acquire the lock here because there is no other task
2255 * which can change it.
2257 cur_trans
->state
= TRANS_STATE_COMPLETED
;
2258 wake_up(&cur_trans
->commit_wait
);
2260 spin_lock(&root
->fs_info
->trans_lock
);
2261 list_del_init(&cur_trans
->list
);
2262 spin_unlock(&root
->fs_info
->trans_lock
);
2264 btrfs_put_transaction(cur_trans
);
2265 btrfs_put_transaction(cur_trans
);
2267 if (trans
->type
& __TRANS_FREEZABLE
)
2268 sb_end_intwrite(root
->fs_info
->sb
);
2270 trace_btrfs_transaction_commit(root
);
2272 btrfs_scrub_continue(root
);
2274 if (current
->journal_info
== trans
)
2275 current
->journal_info
= NULL
;
2277 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
2279 if (current
!= root
->fs_info
->transaction_kthread
&&
2280 current
!= root
->fs_info
->cleaner_kthread
)
2281 btrfs_run_delayed_iputs(root
);
2286 btrfs_scrub_continue(root
);
2287 cleanup_transaction
:
2288 btrfs_trans_release_metadata(trans
, root
);
2289 btrfs_trans_release_chunk_metadata(trans
);
2290 trans
->block_rsv
= NULL
;
2291 btrfs_warn(root
->fs_info
, "Skipping commit of aborted transaction.");
2292 if (current
->journal_info
== trans
)
2293 current
->journal_info
= NULL
;
2294 cleanup_transaction(trans
, root
, ret
);
2300 * return < 0 if error
2301 * 0 if there are no more dead_roots at the time of call
2302 * 1 there are more to be processed, call me again
2304 * The return value indicates there are certainly more snapshots to delete, but
2305 * if there comes a new one during processing, it may return 0. We don't mind,
2306 * because btrfs_commit_super will poke cleaner thread and it will process it a
2307 * few seconds later.
2309 int btrfs_clean_one_deleted_snapshot(struct btrfs_root
*root
)
2312 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2314 spin_lock(&fs_info
->trans_lock
);
2315 if (list_empty(&fs_info
->dead_roots
)) {
2316 spin_unlock(&fs_info
->trans_lock
);
2319 root
= list_first_entry(&fs_info
->dead_roots
,
2320 struct btrfs_root
, root_list
);
2321 list_del_init(&root
->root_list
);
2322 spin_unlock(&fs_info
->trans_lock
);
2324 pr_debug("BTRFS: cleaner removing %llu\n", root
->objectid
);
2326 btrfs_kill_all_delayed_nodes(root
);
2328 if (btrfs_header_backref_rev(root
->node
) <
2329 BTRFS_MIXED_BACKREF_REV
)
2330 ret
= btrfs_drop_snapshot(root
, NULL
, 0, 0);
2332 ret
= btrfs_drop_snapshot(root
, NULL
, 1, 0);
2334 return (ret
< 0) ? 0 : 1;
2337 void btrfs_apply_pending_changes(struct btrfs_fs_info
*fs_info
)
2342 prev
= xchg(&fs_info
->pending_changes
, 0);
2346 bit
= 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE
;
2348 btrfs_set_opt(fs_info
->mount_opt
, INODE_MAP_CACHE
);
2351 bit
= 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE
;
2353 btrfs_clear_opt(fs_info
->mount_opt
, INODE_MAP_CACHE
);
2356 bit
= 1 << BTRFS_PENDING_COMMIT
;
2358 btrfs_debug(fs_info
, "pending commit done");
2363 "unknown pending changes left 0x%lx, ignoring", prev
);