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
);
78 kmem_cache_free(btrfs_transaction_cachep
, transaction
);
82 static void clear_btree_io_tree(struct extent_io_tree
*tree
)
84 spin_lock(&tree
->lock
);
85 while (!RB_EMPTY_ROOT(&tree
->state
)) {
87 struct extent_state
*state
;
89 node
= rb_first(&tree
->state
);
90 state
= rb_entry(node
, struct extent_state
, rb_node
);
91 rb_erase(&state
->rb_node
, &tree
->state
);
92 RB_CLEAR_NODE(&state
->rb_node
);
94 * btree io trees aren't supposed to have tasks waiting for
95 * changes in the flags of extent states ever.
97 ASSERT(!waitqueue_active(&state
->wq
));
98 free_extent_state(state
);
100 cond_resched_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
);
121 /* We can free old roots now. */
122 spin_lock(&trans
->dropped_roots_lock
);
123 while (!list_empty(&trans
->dropped_roots
)) {
124 root
= list_first_entry(&trans
->dropped_roots
,
125 struct btrfs_root
, root_list
);
126 list_del_init(&root
->root_list
);
127 spin_unlock(&trans
->dropped_roots_lock
);
128 btrfs_drop_and_free_fs_root(fs_info
, root
);
129 spin_lock(&trans
->dropped_roots_lock
);
131 spin_unlock(&trans
->dropped_roots_lock
);
132 up_write(&fs_info
->commit_root_sem
);
135 static inline void extwriter_counter_inc(struct btrfs_transaction
*trans
,
138 if (type
& TRANS_EXTWRITERS
)
139 atomic_inc(&trans
->num_extwriters
);
142 static inline void extwriter_counter_dec(struct btrfs_transaction
*trans
,
145 if (type
& TRANS_EXTWRITERS
)
146 atomic_dec(&trans
->num_extwriters
);
149 static inline void extwriter_counter_init(struct btrfs_transaction
*trans
,
152 atomic_set(&trans
->num_extwriters
, ((type
& TRANS_EXTWRITERS
) ? 1 : 0));
155 static inline int extwriter_counter_read(struct btrfs_transaction
*trans
)
157 return atomic_read(&trans
->num_extwriters
);
161 * either allocate a new transaction or hop into the existing one
163 static noinline
int join_transaction(struct btrfs_root
*root
, unsigned int type
)
165 struct btrfs_transaction
*cur_trans
;
166 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
168 spin_lock(&fs_info
->trans_lock
);
170 /* The file system has been taken offline. No new transactions. */
171 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
172 spin_unlock(&fs_info
->trans_lock
);
176 cur_trans
= fs_info
->running_transaction
;
178 if (cur_trans
->aborted
) {
179 spin_unlock(&fs_info
->trans_lock
);
180 return cur_trans
->aborted
;
182 if (btrfs_blocked_trans_types
[cur_trans
->state
] & type
) {
183 spin_unlock(&fs_info
->trans_lock
);
186 atomic_inc(&cur_trans
->use_count
);
187 atomic_inc(&cur_trans
->num_writers
);
188 extwriter_counter_inc(cur_trans
, type
);
189 spin_unlock(&fs_info
->trans_lock
);
192 spin_unlock(&fs_info
->trans_lock
);
195 * If we are ATTACH, we just want to catch the current transaction,
196 * and commit it. If there is no transaction, just return ENOENT.
198 if (type
== TRANS_ATTACH
)
202 * JOIN_NOLOCK only happens during the transaction commit, so
203 * it is impossible that ->running_transaction is NULL
205 BUG_ON(type
== TRANS_JOIN_NOLOCK
);
207 cur_trans
= kmem_cache_alloc(btrfs_transaction_cachep
, GFP_NOFS
);
211 spin_lock(&fs_info
->trans_lock
);
212 if (fs_info
->running_transaction
) {
214 * someone started a transaction after we unlocked. Make sure
215 * to redo the checks above
217 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
219 } else if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
220 spin_unlock(&fs_info
->trans_lock
);
221 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
225 atomic_set(&cur_trans
->num_writers
, 1);
226 extwriter_counter_init(cur_trans
, type
);
227 init_waitqueue_head(&cur_trans
->writer_wait
);
228 init_waitqueue_head(&cur_trans
->commit_wait
);
229 cur_trans
->state
= TRANS_STATE_RUNNING
;
231 * One for this trans handle, one so it will live on until we
232 * commit the transaction.
234 atomic_set(&cur_trans
->use_count
, 2);
235 cur_trans
->have_free_bgs
= 0;
236 cur_trans
->start_time
= get_seconds();
237 cur_trans
->dirty_bg_run
= 0;
239 cur_trans
->delayed_refs
.href_root
= RB_ROOT
;
240 cur_trans
->delayed_refs
.dirty_extent_root
= RB_ROOT
;
241 atomic_set(&cur_trans
->delayed_refs
.num_entries
, 0);
242 cur_trans
->delayed_refs
.num_heads_ready
= 0;
243 cur_trans
->delayed_refs
.pending_csums
= 0;
244 cur_trans
->delayed_refs
.num_heads
= 0;
245 cur_trans
->delayed_refs
.flushing
= 0;
246 cur_trans
->delayed_refs
.run_delayed_start
= 0;
247 cur_trans
->delayed_refs
.qgroup_to_skip
= 0;
250 * although the tree mod log is per file system and not per transaction,
251 * the log must never go across transaction boundaries.
254 if (!list_empty(&fs_info
->tree_mod_seq_list
))
255 WARN(1, KERN_ERR
"BTRFS: tree_mod_seq_list not empty when "
256 "creating a fresh transaction\n");
257 if (!RB_EMPTY_ROOT(&fs_info
->tree_mod_log
))
258 WARN(1, KERN_ERR
"BTRFS: tree_mod_log rb tree not empty when "
259 "creating a fresh transaction\n");
260 atomic64_set(&fs_info
->tree_mod_seq
, 0);
262 spin_lock_init(&cur_trans
->delayed_refs
.lock
);
264 INIT_LIST_HEAD(&cur_trans
->pending_snapshots
);
265 INIT_LIST_HEAD(&cur_trans
->pending_chunks
);
266 INIT_LIST_HEAD(&cur_trans
->switch_commits
);
267 INIT_LIST_HEAD(&cur_trans
->pending_ordered
);
268 INIT_LIST_HEAD(&cur_trans
->dirty_bgs
);
269 INIT_LIST_HEAD(&cur_trans
->io_bgs
);
270 INIT_LIST_HEAD(&cur_trans
->dropped_roots
);
271 mutex_init(&cur_trans
->cache_write_mutex
);
272 cur_trans
->num_dirty_bgs
= 0;
273 spin_lock_init(&cur_trans
->dirty_bgs_lock
);
274 INIT_LIST_HEAD(&cur_trans
->deleted_bgs
);
275 spin_lock_init(&cur_trans
->deleted_bgs_lock
);
276 spin_lock_init(&cur_trans
->dropped_roots_lock
);
277 list_add_tail(&cur_trans
->list
, &fs_info
->trans_list
);
278 extent_io_tree_init(&cur_trans
->dirty_pages
,
279 fs_info
->btree_inode
->i_mapping
);
280 fs_info
->generation
++;
281 cur_trans
->transid
= fs_info
->generation
;
282 fs_info
->running_transaction
= cur_trans
;
283 cur_trans
->aborted
= 0;
284 spin_unlock(&fs_info
->trans_lock
);
290 * this does all the record keeping required to make sure that a reference
291 * counted root is properly recorded in a given transaction. This is required
292 * to make sure the old root from before we joined the transaction is deleted
293 * when the transaction commits
295 static int record_root_in_trans(struct btrfs_trans_handle
*trans
,
296 struct btrfs_root
*root
)
298 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) &&
299 root
->last_trans
< trans
->transid
) {
300 WARN_ON(root
== root
->fs_info
->extent_root
);
301 WARN_ON(root
->commit_root
!= root
->node
);
304 * see below for IN_TRANS_SETUP usage rules
305 * we have the reloc mutex held now, so there
306 * is only one writer in this function
308 set_bit(BTRFS_ROOT_IN_TRANS_SETUP
, &root
->state
);
310 /* make sure readers find IN_TRANS_SETUP before
311 * they find our root->last_trans update
315 spin_lock(&root
->fs_info
->fs_roots_radix_lock
);
316 if (root
->last_trans
== trans
->transid
) {
317 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
320 radix_tree_tag_set(&root
->fs_info
->fs_roots_radix
,
321 (unsigned long)root
->root_key
.objectid
,
322 BTRFS_ROOT_TRANS_TAG
);
323 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
324 root
->last_trans
= trans
->transid
;
326 /* this is pretty tricky. We don't want to
327 * take the relocation lock in btrfs_record_root_in_trans
328 * unless we're really doing the first setup for this root in
331 * Normally we'd use root->last_trans as a flag to decide
332 * if we want to take the expensive mutex.
334 * But, we have to set root->last_trans before we
335 * init the relocation root, otherwise, we trip over warnings
336 * in ctree.c. The solution used here is to flag ourselves
337 * with root IN_TRANS_SETUP. When this is 1, we're still
338 * fixing up the reloc trees and everyone must wait.
340 * When this is zero, they can trust root->last_trans and fly
341 * through btrfs_record_root_in_trans without having to take the
342 * lock. smp_wmb() makes sure that all the writes above are
343 * done before we pop in the zero below
345 btrfs_init_reloc_root(trans
, root
);
346 smp_mb__before_atomic();
347 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP
, &root
->state
);
353 void btrfs_add_dropped_root(struct btrfs_trans_handle
*trans
,
354 struct btrfs_root
*root
)
356 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
358 /* Add ourselves to the transaction dropped list */
359 spin_lock(&cur_trans
->dropped_roots_lock
);
360 list_add_tail(&root
->root_list
, &cur_trans
->dropped_roots
);
361 spin_unlock(&cur_trans
->dropped_roots_lock
);
363 /* Make sure we don't try to update the root at commit time */
364 spin_lock(&root
->fs_info
->fs_roots_radix_lock
);
365 radix_tree_tag_clear(&root
->fs_info
->fs_roots_radix
,
366 (unsigned long)root
->root_key
.objectid
,
367 BTRFS_ROOT_TRANS_TAG
);
368 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
371 int btrfs_record_root_in_trans(struct btrfs_trans_handle
*trans
,
372 struct btrfs_root
*root
)
374 if (!test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
378 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
382 if (root
->last_trans
== trans
->transid
&&
383 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP
, &root
->state
))
386 mutex_lock(&root
->fs_info
->reloc_mutex
);
387 record_root_in_trans(trans
, root
);
388 mutex_unlock(&root
->fs_info
->reloc_mutex
);
393 static inline int is_transaction_blocked(struct btrfs_transaction
*trans
)
395 return (trans
->state
>= TRANS_STATE_BLOCKED
&&
396 trans
->state
< TRANS_STATE_UNBLOCKED
&&
400 /* wait for commit against the current transaction to become unblocked
401 * when this is done, it is safe to start a new transaction, but the current
402 * transaction might not be fully on disk.
404 static void wait_current_trans(struct btrfs_root
*root
)
406 struct btrfs_transaction
*cur_trans
;
408 spin_lock(&root
->fs_info
->trans_lock
);
409 cur_trans
= root
->fs_info
->running_transaction
;
410 if (cur_trans
&& is_transaction_blocked(cur_trans
)) {
411 atomic_inc(&cur_trans
->use_count
);
412 spin_unlock(&root
->fs_info
->trans_lock
);
414 wait_event(root
->fs_info
->transaction_wait
,
415 cur_trans
->state
>= TRANS_STATE_UNBLOCKED
||
417 btrfs_put_transaction(cur_trans
);
419 spin_unlock(&root
->fs_info
->trans_lock
);
423 static int may_wait_transaction(struct btrfs_root
*root
, int type
)
425 if (root
->fs_info
->log_root_recovering
)
428 if (type
== TRANS_USERSPACE
)
431 if (type
== TRANS_START
&&
432 !atomic_read(&root
->fs_info
->open_ioctl_trans
))
438 static inline bool need_reserve_reloc_root(struct btrfs_root
*root
)
440 if (!root
->fs_info
->reloc_ctl
||
441 !test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
442 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
449 static struct btrfs_trans_handle
*
450 start_transaction(struct btrfs_root
*root
, u64 num_items
, unsigned int type
,
451 enum btrfs_reserve_flush_enum flush
)
453 struct btrfs_trans_handle
*h
;
454 struct btrfs_transaction
*cur_trans
;
456 u64 qgroup_reserved
= 0;
457 bool reloc_reserved
= false;
460 /* Send isn't supposed to start transactions. */
461 ASSERT(current
->journal_info
!= BTRFS_SEND_TRANS_STUB
);
463 if (test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
464 return ERR_PTR(-EROFS
);
466 if (current
->journal_info
) {
467 WARN_ON(type
& TRANS_EXTWRITERS
);
468 h
= current
->journal_info
;
470 WARN_ON(h
->use_count
> 2);
471 h
->orig_rsv
= h
->block_rsv
;
477 * Do the reservation before we join the transaction so we can do all
478 * the appropriate flushing if need be.
480 if (num_items
> 0 && root
!= root
->fs_info
->chunk_root
) {
481 if (root
->fs_info
->quota_enabled
&&
482 is_fstree(root
->root_key
.objectid
)) {
483 qgroup_reserved
= num_items
* root
->nodesize
;
484 ret
= btrfs_qgroup_reserve(root
, qgroup_reserved
);
489 num_bytes
= btrfs_calc_trans_metadata_size(root
, num_items
);
491 * Do the reservation for the relocation root creation
493 if (need_reserve_reloc_root(root
)) {
494 num_bytes
+= root
->nodesize
;
495 reloc_reserved
= true;
498 ret
= btrfs_block_rsv_add(root
,
499 &root
->fs_info
->trans_block_rsv
,
505 h
= kmem_cache_alloc(btrfs_trans_handle_cachep
, GFP_NOFS
);
512 * If we are JOIN_NOLOCK we're already committing a transaction and
513 * waiting on this guy, so we don't need to do the sb_start_intwrite
514 * because we're already holding a ref. We need this because we could
515 * have raced in and did an fsync() on a file which can kick a commit
516 * and then we deadlock with somebody doing a freeze.
518 * If we are ATTACH, it means we just want to catch the current
519 * transaction and commit it, so we needn't do sb_start_intwrite().
521 if (type
& __TRANS_FREEZABLE
)
522 sb_start_intwrite(root
->fs_info
->sb
);
524 if (may_wait_transaction(root
, type
))
525 wait_current_trans(root
);
528 ret
= join_transaction(root
, type
);
530 wait_current_trans(root
);
531 if (unlikely(type
== TRANS_ATTACH
))
534 } while (ret
== -EBUSY
);
537 /* We must get the transaction if we are JOIN_NOLOCK. */
538 BUG_ON(type
== TRANS_JOIN_NOLOCK
);
542 cur_trans
= root
->fs_info
->running_transaction
;
544 h
->transid
= cur_trans
->transid
;
545 h
->transaction
= cur_trans
;
547 h
->bytes_reserved
= 0;
548 h
->chunk_bytes_reserved
= 0;
550 h
->delayed_ref_updates
= 0;
556 h
->qgroup_reserved
= 0;
557 h
->delayed_ref_elem
.seq
= 0;
559 h
->allocating_chunk
= false;
560 h
->reloc_reserved
= false;
562 INIT_LIST_HEAD(&h
->qgroup_ref_list
);
563 INIT_LIST_HEAD(&h
->new_bgs
);
564 INIT_LIST_HEAD(&h
->ordered
);
567 if (cur_trans
->state
>= TRANS_STATE_BLOCKED
&&
568 may_wait_transaction(root
, type
)) {
569 current
->journal_info
= h
;
570 btrfs_commit_transaction(h
, root
);
575 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
576 h
->transid
, num_bytes
, 1);
577 h
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
578 h
->bytes_reserved
= num_bytes
;
579 h
->reloc_reserved
= reloc_reserved
;
581 h
->qgroup_reserved
= qgroup_reserved
;
584 btrfs_record_root_in_trans(h
, root
);
586 if (!current
->journal_info
&& type
!= TRANS_USERSPACE
)
587 current
->journal_info
= h
;
591 if (type
& __TRANS_FREEZABLE
)
592 sb_end_intwrite(root
->fs_info
->sb
);
593 kmem_cache_free(btrfs_trans_handle_cachep
, h
);
596 btrfs_block_rsv_release(root
, &root
->fs_info
->trans_block_rsv
,
600 btrfs_qgroup_free(root
, qgroup_reserved
);
604 struct btrfs_trans_handle
*btrfs_start_transaction(struct btrfs_root
*root
,
607 return start_transaction(root
, num_items
, TRANS_START
,
608 BTRFS_RESERVE_FLUSH_ALL
);
611 struct btrfs_trans_handle
*btrfs_start_transaction_lflush(
612 struct btrfs_root
*root
, int num_items
)
614 return start_transaction(root
, num_items
, TRANS_START
,
615 BTRFS_RESERVE_FLUSH_LIMIT
);
618 struct btrfs_trans_handle
*btrfs_join_transaction(struct btrfs_root
*root
)
620 return start_transaction(root
, 0, TRANS_JOIN
, 0);
623 struct btrfs_trans_handle
*btrfs_join_transaction_nolock(struct btrfs_root
*root
)
625 return start_transaction(root
, 0, TRANS_JOIN_NOLOCK
, 0);
628 struct btrfs_trans_handle
*btrfs_start_ioctl_transaction(struct btrfs_root
*root
)
630 return start_transaction(root
, 0, TRANS_USERSPACE
, 0);
634 * btrfs_attach_transaction() - catch the running transaction
636 * It is used when we want to commit the current the transaction, but
637 * don't want to start a new one.
639 * Note: If this function return -ENOENT, it just means there is no
640 * running transaction. But it is possible that the inactive transaction
641 * is still in the memory, not fully on disk. If you hope there is no
642 * inactive transaction in the fs when -ENOENT is returned, you should
644 * btrfs_attach_transaction_barrier()
646 struct btrfs_trans_handle
*btrfs_attach_transaction(struct btrfs_root
*root
)
648 return start_transaction(root
, 0, TRANS_ATTACH
, 0);
652 * btrfs_attach_transaction_barrier() - catch the running transaction
654 * It is similar to the above function, the differentia is this one
655 * will wait for all the inactive transactions until they fully
658 struct btrfs_trans_handle
*
659 btrfs_attach_transaction_barrier(struct btrfs_root
*root
)
661 struct btrfs_trans_handle
*trans
;
663 trans
= start_transaction(root
, 0, TRANS_ATTACH
, 0);
664 if (IS_ERR(trans
) && PTR_ERR(trans
) == -ENOENT
)
665 btrfs_wait_for_commit(root
, 0);
670 /* wait for a transaction commit to be fully complete */
671 static noinline
void wait_for_commit(struct btrfs_root
*root
,
672 struct btrfs_transaction
*commit
)
674 wait_event(commit
->commit_wait
, commit
->state
== TRANS_STATE_COMPLETED
);
677 int btrfs_wait_for_commit(struct btrfs_root
*root
, u64 transid
)
679 struct btrfs_transaction
*cur_trans
= NULL
, *t
;
683 if (transid
<= root
->fs_info
->last_trans_committed
)
686 /* find specified transaction */
687 spin_lock(&root
->fs_info
->trans_lock
);
688 list_for_each_entry(t
, &root
->fs_info
->trans_list
, list
) {
689 if (t
->transid
== transid
) {
691 atomic_inc(&cur_trans
->use_count
);
695 if (t
->transid
> transid
) {
700 spin_unlock(&root
->fs_info
->trans_lock
);
703 * The specified transaction doesn't exist, or we
704 * raced with btrfs_commit_transaction
707 if (transid
> root
->fs_info
->last_trans_committed
)
712 /* find newest transaction that is committing | committed */
713 spin_lock(&root
->fs_info
->trans_lock
);
714 list_for_each_entry_reverse(t
, &root
->fs_info
->trans_list
,
716 if (t
->state
>= TRANS_STATE_COMMIT_START
) {
717 if (t
->state
== TRANS_STATE_COMPLETED
)
720 atomic_inc(&cur_trans
->use_count
);
724 spin_unlock(&root
->fs_info
->trans_lock
);
726 goto out
; /* nothing committing|committed */
729 wait_for_commit(root
, cur_trans
);
730 btrfs_put_transaction(cur_trans
);
735 void btrfs_throttle(struct btrfs_root
*root
)
737 if (!atomic_read(&root
->fs_info
->open_ioctl_trans
))
738 wait_current_trans(root
);
741 static int should_end_transaction(struct btrfs_trans_handle
*trans
,
742 struct btrfs_root
*root
)
744 if (root
->fs_info
->global_block_rsv
.space_info
->full
&&
745 btrfs_check_space_for_delayed_refs(trans
, root
))
748 return !!btrfs_block_rsv_check(root
, &root
->fs_info
->global_block_rsv
, 5);
751 int btrfs_should_end_transaction(struct btrfs_trans_handle
*trans
,
752 struct btrfs_root
*root
)
754 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
759 if (cur_trans
->state
>= TRANS_STATE_BLOCKED
||
760 cur_trans
->delayed_refs
.flushing
)
763 updates
= trans
->delayed_ref_updates
;
764 trans
->delayed_ref_updates
= 0;
766 err
= btrfs_run_delayed_refs(trans
, root
, updates
* 2);
767 if (err
) /* Error code will also eval true */
771 return should_end_transaction(trans
, root
);
774 static int __btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
775 struct btrfs_root
*root
, int throttle
)
777 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
778 struct btrfs_fs_info
*info
= root
->fs_info
;
779 unsigned long cur
= trans
->delayed_ref_updates
;
780 int lock
= (trans
->type
!= TRANS_JOIN_NOLOCK
);
782 int must_run_delayed_refs
= 0;
784 if (trans
->use_count
> 1) {
786 trans
->block_rsv
= trans
->orig_rsv
;
790 btrfs_trans_release_metadata(trans
, root
);
791 trans
->block_rsv
= NULL
;
793 if (!list_empty(&trans
->new_bgs
))
794 btrfs_create_pending_block_groups(trans
, root
);
796 if (!list_empty(&trans
->ordered
)) {
797 spin_lock(&info
->trans_lock
);
798 list_splice_init(&trans
->ordered
, &cur_trans
->pending_ordered
);
799 spin_unlock(&info
->trans_lock
);
802 trans
->delayed_ref_updates
= 0;
804 must_run_delayed_refs
=
805 btrfs_should_throttle_delayed_refs(trans
, root
);
806 cur
= max_t(unsigned long, cur
, 32);
809 * don't make the caller wait if they are from a NOLOCK
810 * or ATTACH transaction, it will deadlock with commit
812 if (must_run_delayed_refs
== 1 &&
813 (trans
->type
& (__TRANS_JOIN_NOLOCK
| __TRANS_ATTACH
)))
814 must_run_delayed_refs
= 2;
817 if (trans
->qgroup_reserved
) {
819 * the same root has to be passed here between start_transaction
820 * and end_transaction. Subvolume quota depends on this.
822 btrfs_qgroup_free(trans
->root
, trans
->qgroup_reserved
);
823 trans
->qgroup_reserved
= 0;
826 btrfs_trans_release_metadata(trans
, root
);
827 trans
->block_rsv
= NULL
;
829 if (!list_empty(&trans
->new_bgs
))
830 btrfs_create_pending_block_groups(trans
, root
);
832 btrfs_trans_release_chunk_metadata(trans
);
834 if (lock
&& !atomic_read(&root
->fs_info
->open_ioctl_trans
) &&
835 should_end_transaction(trans
, root
) &&
836 ACCESS_ONCE(cur_trans
->state
) == TRANS_STATE_RUNNING
) {
837 spin_lock(&info
->trans_lock
);
838 if (cur_trans
->state
== TRANS_STATE_RUNNING
)
839 cur_trans
->state
= TRANS_STATE_BLOCKED
;
840 spin_unlock(&info
->trans_lock
);
843 if (lock
&& ACCESS_ONCE(cur_trans
->state
) == TRANS_STATE_BLOCKED
) {
845 return btrfs_commit_transaction(trans
, root
);
847 wake_up_process(info
->transaction_kthread
);
850 if (trans
->type
& __TRANS_FREEZABLE
)
851 sb_end_intwrite(root
->fs_info
->sb
);
853 WARN_ON(cur_trans
!= info
->running_transaction
);
854 WARN_ON(atomic_read(&cur_trans
->num_writers
) < 1);
855 atomic_dec(&cur_trans
->num_writers
);
856 extwriter_counter_dec(cur_trans
, trans
->type
);
859 if (waitqueue_active(&cur_trans
->writer_wait
))
860 wake_up(&cur_trans
->writer_wait
);
861 btrfs_put_transaction(cur_trans
);
863 if (current
->journal_info
== trans
)
864 current
->journal_info
= NULL
;
867 btrfs_run_delayed_iputs(root
);
869 if (trans
->aborted
||
870 test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
)) {
871 wake_up_process(info
->transaction_kthread
);
874 assert_qgroups_uptodate(trans
);
876 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
877 if (must_run_delayed_refs
) {
878 btrfs_async_run_delayed_refs(root
, cur
,
879 must_run_delayed_refs
== 1);
884 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
885 struct btrfs_root
*root
)
887 return __btrfs_end_transaction(trans
, root
, 0);
890 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
,
891 struct btrfs_root
*root
)
893 return __btrfs_end_transaction(trans
, root
, 1);
897 * when btree blocks are allocated, they have some corresponding bits set for
898 * them in one of two extent_io trees. This is used to make sure all of
899 * those extents are sent to disk but does not wait on them
901 int btrfs_write_marked_extents(struct btrfs_root
*root
,
902 struct extent_io_tree
*dirty_pages
, int mark
)
906 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
907 struct extent_state
*cached_state
= NULL
;
911 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
912 mark
, &cached_state
)) {
913 bool wait_writeback
= false;
915 err
= convert_extent_bit(dirty_pages
, start
, end
,
917 mark
, &cached_state
, GFP_NOFS
);
919 * convert_extent_bit can return -ENOMEM, which is most of the
920 * time a temporary error. So when it happens, ignore the error
921 * and wait for writeback of this range to finish - because we
922 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
923 * to btrfs_wait_marked_extents() would not know that writeback
924 * for this range started and therefore wouldn't wait for it to
925 * finish - we don't want to commit a superblock that points to
926 * btree nodes/leafs for which writeback hasn't finished yet
927 * (and without errors).
928 * We cleanup any entries left in the io tree when committing
929 * the transaction (through clear_btree_io_tree()).
931 if (err
== -ENOMEM
) {
933 wait_writeback
= true;
936 err
= filemap_fdatawrite_range(mapping
, start
, end
);
939 else if (wait_writeback
)
940 werr
= filemap_fdatawait_range(mapping
, start
, end
);
941 free_extent_state(cached_state
);
950 * when btree blocks are allocated, they have some corresponding bits set for
951 * them in one of two extent_io trees. This is used to make sure all of
952 * those extents are on disk for transaction or log commit. We wait
953 * on all the pages and clear them from the dirty pages state tree
955 int btrfs_wait_marked_extents(struct btrfs_root
*root
,
956 struct extent_io_tree
*dirty_pages
, int mark
)
960 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
961 struct extent_state
*cached_state
= NULL
;
964 struct btrfs_inode
*btree_ino
= BTRFS_I(root
->fs_info
->btree_inode
);
967 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
968 EXTENT_NEED_WAIT
, &cached_state
)) {
970 * Ignore -ENOMEM errors returned by clear_extent_bit().
971 * When committing the transaction, we'll remove any entries
972 * left in the io tree. For a log commit, we don't remove them
973 * after committing the log because the tree can be accessed
974 * concurrently - we do it only at transaction commit time when
975 * it's safe to do it (through clear_btree_io_tree()).
977 err
= clear_extent_bit(dirty_pages
, start
, end
,
979 0, 0, &cached_state
, GFP_NOFS
);
983 err
= filemap_fdatawait_range(mapping
, start
, end
);
986 free_extent_state(cached_state
);
994 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
995 if ((mark
& EXTENT_DIRTY
) &&
996 test_and_clear_bit(BTRFS_INODE_BTREE_LOG1_ERR
,
997 &btree_ino
->runtime_flags
))
1000 if ((mark
& EXTENT_NEW
) &&
1001 test_and_clear_bit(BTRFS_INODE_BTREE_LOG2_ERR
,
1002 &btree_ino
->runtime_flags
))
1005 if (test_and_clear_bit(BTRFS_INODE_BTREE_ERR
,
1006 &btree_ino
->runtime_flags
))
1010 if (errors
&& !werr
)
1017 * when btree blocks are allocated, they have some corresponding bits set for
1018 * them in one of two extent_io trees. This is used to make sure all of
1019 * those extents are on disk for transaction or log commit
1021 static int btrfs_write_and_wait_marked_extents(struct btrfs_root
*root
,
1022 struct extent_io_tree
*dirty_pages
, int mark
)
1026 struct blk_plug plug
;
1028 blk_start_plug(&plug
);
1029 ret
= btrfs_write_marked_extents(root
, dirty_pages
, mark
);
1030 blk_finish_plug(&plug
);
1031 ret2
= btrfs_wait_marked_extents(root
, dirty_pages
, mark
);
1040 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
,
1041 struct btrfs_root
*root
)
1045 ret
= btrfs_write_and_wait_marked_extents(root
,
1046 &trans
->transaction
->dirty_pages
,
1048 clear_btree_io_tree(&trans
->transaction
->dirty_pages
);
1054 * this is used to update the root pointer in the tree of tree roots.
1056 * But, in the case of the extent allocation tree, updating the root
1057 * pointer may allocate blocks which may change the root of the extent
1060 * So, this loops and repeats and makes sure the cowonly root didn't
1061 * change while the root pointer was being updated in the metadata.
1063 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
1064 struct btrfs_root
*root
)
1067 u64 old_root_bytenr
;
1069 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
1071 old_root_used
= btrfs_root_used(&root
->root_item
);
1074 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
1075 if (old_root_bytenr
== root
->node
->start
&&
1076 old_root_used
== btrfs_root_used(&root
->root_item
))
1079 btrfs_set_root_node(&root
->root_item
, root
->node
);
1080 ret
= btrfs_update_root(trans
, tree_root
,
1086 old_root_used
= btrfs_root_used(&root
->root_item
);
1093 * update all the cowonly tree roots on disk
1095 * The error handling in this function may not be obvious. Any of the
1096 * failures will cause the file system to go offline. We still need
1097 * to clean up the delayed refs.
1099 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
,
1100 struct btrfs_root
*root
)
1102 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1103 struct list_head
*dirty_bgs
= &trans
->transaction
->dirty_bgs
;
1104 struct list_head
*io_bgs
= &trans
->transaction
->io_bgs
;
1105 struct list_head
*next
;
1106 struct extent_buffer
*eb
;
1109 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
1110 ret
= btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
,
1112 btrfs_tree_unlock(eb
);
1113 free_extent_buffer(eb
);
1118 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1122 ret
= btrfs_run_dev_stats(trans
, root
->fs_info
);
1125 ret
= btrfs_run_dev_replace(trans
, root
->fs_info
);
1128 ret
= btrfs_run_qgroups(trans
, root
->fs_info
);
1132 ret
= btrfs_setup_space_cache(trans
, root
);
1136 /* run_qgroups might have added some more refs */
1137 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1141 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
1142 next
= fs_info
->dirty_cowonly_roots
.next
;
1143 list_del_init(next
);
1144 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
1145 clear_bit(BTRFS_ROOT_DIRTY
, &root
->state
);
1147 if (root
!= fs_info
->extent_root
)
1148 list_add_tail(&root
->dirty_list
,
1149 &trans
->transaction
->switch_commits
);
1150 ret
= update_cowonly_root(trans
, root
);
1153 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1158 while (!list_empty(dirty_bgs
) || !list_empty(io_bgs
)) {
1159 ret
= btrfs_write_dirty_block_groups(trans
, root
);
1162 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1167 if (!list_empty(&fs_info
->dirty_cowonly_roots
))
1170 list_add_tail(&fs_info
->extent_root
->dirty_list
,
1171 &trans
->transaction
->switch_commits
);
1172 btrfs_after_dev_replace_commit(fs_info
);
1178 * dead roots are old snapshots that need to be deleted. This allocates
1179 * a dirty root struct and adds it into the list of dead roots that need to
1182 void btrfs_add_dead_root(struct btrfs_root
*root
)
1184 spin_lock(&root
->fs_info
->trans_lock
);
1185 if (list_empty(&root
->root_list
))
1186 list_add_tail(&root
->root_list
, &root
->fs_info
->dead_roots
);
1187 spin_unlock(&root
->fs_info
->trans_lock
);
1191 * update all the cowonly tree roots on disk
1193 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
,
1194 struct btrfs_root
*root
)
1196 struct btrfs_root
*gang
[8];
1197 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1202 spin_lock(&fs_info
->fs_roots_radix_lock
);
1204 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
1207 BTRFS_ROOT_TRANS_TAG
);
1210 for (i
= 0; i
< ret
; i
++) {
1212 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
1213 (unsigned long)root
->root_key
.objectid
,
1214 BTRFS_ROOT_TRANS_TAG
);
1215 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1217 btrfs_free_log(trans
, root
);
1218 btrfs_update_reloc_root(trans
, root
);
1219 btrfs_orphan_commit_root(trans
, root
);
1221 btrfs_save_ino_cache(root
, trans
);
1223 /* see comments in should_cow_block() */
1224 clear_bit(BTRFS_ROOT_FORCE_COW
, &root
->state
);
1225 smp_mb__after_atomic();
1227 if (root
->commit_root
!= root
->node
) {
1228 list_add_tail(&root
->dirty_list
,
1229 &trans
->transaction
->switch_commits
);
1230 btrfs_set_root_node(&root
->root_item
,
1234 err
= btrfs_update_root(trans
, fs_info
->tree_root
,
1237 spin_lock(&fs_info
->fs_roots_radix_lock
);
1242 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1247 * defrag a given btree.
1248 * Every leaf in the btree is read and defragged.
1250 int btrfs_defrag_root(struct btrfs_root
*root
)
1252 struct btrfs_fs_info
*info
= root
->fs_info
;
1253 struct btrfs_trans_handle
*trans
;
1256 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING
, &root
->state
))
1260 trans
= btrfs_start_transaction(root
, 0);
1262 return PTR_ERR(trans
);
1264 ret
= btrfs_defrag_leaves(trans
, root
);
1266 btrfs_end_transaction(trans
, root
);
1267 btrfs_btree_balance_dirty(info
->tree_root
);
1270 if (btrfs_fs_closing(root
->fs_info
) || ret
!= -EAGAIN
)
1273 if (btrfs_defrag_cancelled(root
->fs_info
)) {
1274 pr_debug("BTRFS: defrag_root cancelled\n");
1279 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING
, &root
->state
);
1284 * new snapshots need to be created at a very specific time in the
1285 * transaction commit. This does the actual creation.
1288 * If the error which may affect the commitment of the current transaction
1289 * happens, we should return the error number. If the error which just affect
1290 * the creation of the pending snapshots, just return 0.
1292 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
1293 struct btrfs_fs_info
*fs_info
,
1294 struct btrfs_pending_snapshot
*pending
)
1296 struct btrfs_key key
;
1297 struct btrfs_root_item
*new_root_item
;
1298 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1299 struct btrfs_root
*root
= pending
->root
;
1300 struct btrfs_root
*parent_root
;
1301 struct btrfs_block_rsv
*rsv
;
1302 struct inode
*parent_inode
;
1303 struct btrfs_path
*path
;
1304 struct btrfs_dir_item
*dir_item
;
1305 struct dentry
*dentry
;
1306 struct extent_buffer
*tmp
;
1307 struct extent_buffer
*old
;
1308 struct timespec cur_time
= CURRENT_TIME
;
1316 path
= btrfs_alloc_path();
1318 pending
->error
= -ENOMEM
;
1322 new_root_item
= kmalloc(sizeof(*new_root_item
), GFP_NOFS
);
1323 if (!new_root_item
) {
1324 pending
->error
= -ENOMEM
;
1325 goto root_item_alloc_fail
;
1328 pending
->error
= btrfs_find_free_objectid(tree_root
, &objectid
);
1330 goto no_free_objectid
;
1333 * Make qgroup to skip current new snapshot's qgroupid, as it is
1334 * accounted by later btrfs_qgroup_inherit().
1336 btrfs_set_skip_qgroup(trans
, objectid
);
1338 btrfs_reloc_pre_snapshot(pending
, &to_reserve
);
1340 if (to_reserve
> 0) {
1341 pending
->error
= btrfs_block_rsv_add(root
,
1342 &pending
->block_rsv
,
1344 BTRFS_RESERVE_NO_FLUSH
);
1346 goto clear_skip_qgroup
;
1349 key
.objectid
= objectid
;
1350 key
.offset
= (u64
)-1;
1351 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1353 rsv
= trans
->block_rsv
;
1354 trans
->block_rsv
= &pending
->block_rsv
;
1355 trans
->bytes_reserved
= trans
->block_rsv
->reserved
;
1357 dentry
= pending
->dentry
;
1358 parent_inode
= pending
->dir
;
1359 parent_root
= BTRFS_I(parent_inode
)->root
;
1360 record_root_in_trans(trans
, parent_root
);
1363 * insert the directory item
1365 ret
= btrfs_set_inode_index(parent_inode
, &index
);
1366 BUG_ON(ret
); /* -ENOMEM */
1368 /* check if there is a file/dir which has the same name. */
1369 dir_item
= btrfs_lookup_dir_item(NULL
, parent_root
, path
,
1370 btrfs_ino(parent_inode
),
1371 dentry
->d_name
.name
,
1372 dentry
->d_name
.len
, 0);
1373 if (dir_item
!= NULL
&& !IS_ERR(dir_item
)) {
1374 pending
->error
= -EEXIST
;
1375 goto dir_item_existed
;
1376 } else if (IS_ERR(dir_item
)) {
1377 ret
= PTR_ERR(dir_item
);
1378 btrfs_abort_transaction(trans
, root
, ret
);
1381 btrfs_release_path(path
);
1384 * pull in the delayed directory update
1385 * and the delayed inode item
1386 * otherwise we corrupt the FS during
1389 ret
= btrfs_run_delayed_items(trans
, root
);
1390 if (ret
) { /* Transaction aborted */
1391 btrfs_abort_transaction(trans
, root
, ret
);
1395 record_root_in_trans(trans
, root
);
1396 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
1397 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
1398 btrfs_check_and_init_root_item(new_root_item
);
1400 root_flags
= btrfs_root_flags(new_root_item
);
1401 if (pending
->readonly
)
1402 root_flags
|= BTRFS_ROOT_SUBVOL_RDONLY
;
1404 root_flags
&= ~BTRFS_ROOT_SUBVOL_RDONLY
;
1405 btrfs_set_root_flags(new_root_item
, root_flags
);
1407 btrfs_set_root_generation_v2(new_root_item
,
1409 uuid_le_gen(&new_uuid
);
1410 memcpy(new_root_item
->uuid
, new_uuid
.b
, BTRFS_UUID_SIZE
);
1411 memcpy(new_root_item
->parent_uuid
, root
->root_item
.uuid
,
1413 if (!(root_flags
& BTRFS_ROOT_SUBVOL_RDONLY
)) {
1414 memset(new_root_item
->received_uuid
, 0,
1415 sizeof(new_root_item
->received_uuid
));
1416 memset(&new_root_item
->stime
, 0, sizeof(new_root_item
->stime
));
1417 memset(&new_root_item
->rtime
, 0, sizeof(new_root_item
->rtime
));
1418 btrfs_set_root_stransid(new_root_item
, 0);
1419 btrfs_set_root_rtransid(new_root_item
, 0);
1421 btrfs_set_stack_timespec_sec(&new_root_item
->otime
, cur_time
.tv_sec
);
1422 btrfs_set_stack_timespec_nsec(&new_root_item
->otime
, cur_time
.tv_nsec
);
1423 btrfs_set_root_otransid(new_root_item
, trans
->transid
);
1425 old
= btrfs_lock_root_node(root
);
1426 ret
= btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
);
1428 btrfs_tree_unlock(old
);
1429 free_extent_buffer(old
);
1430 btrfs_abort_transaction(trans
, root
, ret
);
1434 btrfs_set_lock_blocking(old
);
1436 ret
= btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
1437 /* clean up in any case */
1438 btrfs_tree_unlock(old
);
1439 free_extent_buffer(old
);
1441 btrfs_abort_transaction(trans
, root
, ret
);
1444 /* see comments in should_cow_block() */
1445 set_bit(BTRFS_ROOT_FORCE_COW
, &root
->state
);
1448 btrfs_set_root_node(new_root_item
, tmp
);
1449 /* record when the snapshot was created in key.offset */
1450 key
.offset
= trans
->transid
;
1451 ret
= btrfs_insert_root(trans
, tree_root
, &key
, new_root_item
);
1452 btrfs_tree_unlock(tmp
);
1453 free_extent_buffer(tmp
);
1455 btrfs_abort_transaction(trans
, root
, ret
);
1460 * insert root back/forward references
1462 ret
= btrfs_add_root_ref(trans
, tree_root
, objectid
,
1463 parent_root
->root_key
.objectid
,
1464 btrfs_ino(parent_inode
), index
,
1465 dentry
->d_name
.name
, dentry
->d_name
.len
);
1467 btrfs_abort_transaction(trans
, root
, ret
);
1471 key
.offset
= (u64
)-1;
1472 pending
->snap
= btrfs_read_fs_root_no_name(root
->fs_info
, &key
);
1473 if (IS_ERR(pending
->snap
)) {
1474 ret
= PTR_ERR(pending
->snap
);
1475 btrfs_abort_transaction(trans
, root
, ret
);
1479 ret
= btrfs_reloc_post_snapshot(trans
, pending
);
1481 btrfs_abort_transaction(trans
, root
, ret
);
1485 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1487 btrfs_abort_transaction(trans
, root
, ret
);
1491 ret
= btrfs_insert_dir_item(trans
, parent_root
,
1492 dentry
->d_name
.name
, dentry
->d_name
.len
,
1494 BTRFS_FT_DIR
, index
);
1495 /* We have check then name at the beginning, so it is impossible. */
1496 BUG_ON(ret
== -EEXIST
|| ret
== -EOVERFLOW
);
1498 btrfs_abort_transaction(trans
, root
, ret
);
1502 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
1503 dentry
->d_name
.len
* 2);
1504 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
1505 ret
= btrfs_update_inode_fallback(trans
, parent_root
, parent_inode
);
1507 btrfs_abort_transaction(trans
, root
, ret
);
1510 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
, new_uuid
.b
,
1511 BTRFS_UUID_KEY_SUBVOL
, objectid
);
1513 btrfs_abort_transaction(trans
, root
, ret
);
1516 if (!btrfs_is_empty_uuid(new_root_item
->received_uuid
)) {
1517 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
1518 new_root_item
->received_uuid
,
1519 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
1521 if (ret
&& ret
!= -EEXIST
) {
1522 btrfs_abort_transaction(trans
, root
, ret
);
1527 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1529 btrfs_abort_transaction(trans
, root
, ret
);
1534 * account qgroup counters before qgroup_inherit()
1536 ret
= btrfs_qgroup_prepare_account_extents(trans
, fs_info
);
1539 ret
= btrfs_qgroup_account_extents(trans
, fs_info
);
1542 ret
= btrfs_qgroup_inherit(trans
, fs_info
,
1543 root
->root_key
.objectid
,
1544 objectid
, pending
->inherit
);
1546 btrfs_abort_transaction(trans
, root
, ret
);
1551 pending
->error
= ret
;
1553 trans
->block_rsv
= rsv
;
1554 trans
->bytes_reserved
= 0;
1556 btrfs_clear_skip_qgroup(trans
);
1558 kfree(new_root_item
);
1559 root_item_alloc_fail
:
1560 btrfs_free_path(path
);
1565 * create all the snapshots we've scheduled for creation
1567 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
,
1568 struct btrfs_fs_info
*fs_info
)
1570 struct btrfs_pending_snapshot
*pending
, *next
;
1571 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
1574 list_for_each_entry_safe(pending
, next
, head
, list
) {
1575 list_del(&pending
->list
);
1576 ret
= create_pending_snapshot(trans
, fs_info
, pending
);
1583 static void update_super_roots(struct btrfs_root
*root
)
1585 struct btrfs_root_item
*root_item
;
1586 struct btrfs_super_block
*super
;
1588 super
= root
->fs_info
->super_copy
;
1590 root_item
= &root
->fs_info
->chunk_root
->root_item
;
1591 super
->chunk_root
= root_item
->bytenr
;
1592 super
->chunk_root_generation
= root_item
->generation
;
1593 super
->chunk_root_level
= root_item
->level
;
1595 root_item
= &root
->fs_info
->tree_root
->root_item
;
1596 super
->root
= root_item
->bytenr
;
1597 super
->generation
= root_item
->generation
;
1598 super
->root_level
= root_item
->level
;
1599 if (btrfs_test_opt(root
, SPACE_CACHE
))
1600 super
->cache_generation
= root_item
->generation
;
1601 if (root
->fs_info
->update_uuid_tree_gen
)
1602 super
->uuid_tree_generation
= root_item
->generation
;
1605 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
1607 struct btrfs_transaction
*trans
;
1610 spin_lock(&info
->trans_lock
);
1611 trans
= info
->running_transaction
;
1613 ret
= (trans
->state
>= TRANS_STATE_COMMIT_START
);
1614 spin_unlock(&info
->trans_lock
);
1618 int btrfs_transaction_blocked(struct btrfs_fs_info
*info
)
1620 struct btrfs_transaction
*trans
;
1623 spin_lock(&info
->trans_lock
);
1624 trans
= info
->running_transaction
;
1626 ret
= is_transaction_blocked(trans
);
1627 spin_unlock(&info
->trans_lock
);
1632 * wait for the current transaction commit to start and block subsequent
1635 static void wait_current_trans_commit_start(struct btrfs_root
*root
,
1636 struct btrfs_transaction
*trans
)
1638 wait_event(root
->fs_info
->transaction_blocked_wait
,
1639 trans
->state
>= TRANS_STATE_COMMIT_START
||
1644 * wait for the current transaction to start and then become unblocked.
1647 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root
*root
,
1648 struct btrfs_transaction
*trans
)
1650 wait_event(root
->fs_info
->transaction_wait
,
1651 trans
->state
>= TRANS_STATE_UNBLOCKED
||
1656 * commit transactions asynchronously. once btrfs_commit_transaction_async
1657 * returns, any subsequent transaction will not be allowed to join.
1659 struct btrfs_async_commit
{
1660 struct btrfs_trans_handle
*newtrans
;
1661 struct btrfs_root
*root
;
1662 struct work_struct work
;
1665 static void do_async_commit(struct work_struct
*work
)
1667 struct btrfs_async_commit
*ac
=
1668 container_of(work
, struct btrfs_async_commit
, work
);
1671 * We've got freeze protection passed with the transaction.
1672 * Tell lockdep about it.
1674 if (ac
->newtrans
->type
& __TRANS_FREEZABLE
)
1675 __sb_writers_acquired(ac
->root
->fs_info
->sb
, SB_FREEZE_FS
);
1677 current
->journal_info
= ac
->newtrans
;
1679 btrfs_commit_transaction(ac
->newtrans
, ac
->root
);
1683 int btrfs_commit_transaction_async(struct btrfs_trans_handle
*trans
,
1684 struct btrfs_root
*root
,
1685 int wait_for_unblock
)
1687 struct btrfs_async_commit
*ac
;
1688 struct btrfs_transaction
*cur_trans
;
1690 ac
= kmalloc(sizeof(*ac
), GFP_NOFS
);
1694 INIT_WORK(&ac
->work
, do_async_commit
);
1696 ac
->newtrans
= btrfs_join_transaction(root
);
1697 if (IS_ERR(ac
->newtrans
)) {
1698 int err
= PTR_ERR(ac
->newtrans
);
1703 /* take transaction reference */
1704 cur_trans
= trans
->transaction
;
1705 atomic_inc(&cur_trans
->use_count
);
1707 btrfs_end_transaction(trans
, root
);
1710 * Tell lockdep we've released the freeze rwsem, since the
1711 * async commit thread will be the one to unlock it.
1713 if (ac
->newtrans
->type
& __TRANS_FREEZABLE
)
1714 __sb_writers_release(root
->fs_info
->sb
, SB_FREEZE_FS
);
1716 schedule_work(&ac
->work
);
1718 /* wait for transaction to start and unblock */
1719 if (wait_for_unblock
)
1720 wait_current_trans_commit_start_and_unblock(root
, cur_trans
);
1722 wait_current_trans_commit_start(root
, cur_trans
);
1724 if (current
->journal_info
== trans
)
1725 current
->journal_info
= NULL
;
1727 btrfs_put_transaction(cur_trans
);
1732 static void cleanup_transaction(struct btrfs_trans_handle
*trans
,
1733 struct btrfs_root
*root
, int err
)
1735 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1738 WARN_ON(trans
->use_count
> 1);
1740 btrfs_abort_transaction(trans
, root
, err
);
1742 spin_lock(&root
->fs_info
->trans_lock
);
1745 * If the transaction is removed from the list, it means this
1746 * transaction has been committed successfully, so it is impossible
1747 * to call the cleanup function.
1749 BUG_ON(list_empty(&cur_trans
->list
));
1751 list_del_init(&cur_trans
->list
);
1752 if (cur_trans
== root
->fs_info
->running_transaction
) {
1753 cur_trans
->state
= TRANS_STATE_COMMIT_DOING
;
1754 spin_unlock(&root
->fs_info
->trans_lock
);
1755 wait_event(cur_trans
->writer_wait
,
1756 atomic_read(&cur_trans
->num_writers
) == 1);
1758 spin_lock(&root
->fs_info
->trans_lock
);
1760 spin_unlock(&root
->fs_info
->trans_lock
);
1762 btrfs_cleanup_one_transaction(trans
->transaction
, root
);
1764 spin_lock(&root
->fs_info
->trans_lock
);
1765 if (cur_trans
== root
->fs_info
->running_transaction
)
1766 root
->fs_info
->running_transaction
= NULL
;
1767 spin_unlock(&root
->fs_info
->trans_lock
);
1769 if (trans
->type
& __TRANS_FREEZABLE
)
1770 sb_end_intwrite(root
->fs_info
->sb
);
1771 btrfs_put_transaction(cur_trans
);
1772 btrfs_put_transaction(cur_trans
);
1774 trace_btrfs_transaction_commit(root
);
1776 if (current
->journal_info
== trans
)
1777 current
->journal_info
= NULL
;
1778 btrfs_scrub_cancel(root
->fs_info
);
1780 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1783 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info
*fs_info
)
1785 if (btrfs_test_opt(fs_info
->tree_root
, FLUSHONCOMMIT
))
1786 return btrfs_start_delalloc_roots(fs_info
, 1, -1);
1790 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info
*fs_info
)
1792 if (btrfs_test_opt(fs_info
->tree_root
, FLUSHONCOMMIT
))
1793 btrfs_wait_ordered_roots(fs_info
, -1);
1797 btrfs_wait_pending_ordered(struct btrfs_transaction
*cur_trans
,
1798 struct btrfs_fs_info
*fs_info
)
1800 struct btrfs_ordered_extent
*ordered
;
1802 spin_lock(&fs_info
->trans_lock
);
1803 while (!list_empty(&cur_trans
->pending_ordered
)) {
1804 ordered
= list_first_entry(&cur_trans
->pending_ordered
,
1805 struct btrfs_ordered_extent
,
1807 list_del_init(&ordered
->trans_list
);
1808 spin_unlock(&fs_info
->trans_lock
);
1810 wait_event(ordered
->wait
, test_bit(BTRFS_ORDERED_COMPLETE
,
1812 btrfs_put_ordered_extent(ordered
);
1813 spin_lock(&fs_info
->trans_lock
);
1815 spin_unlock(&fs_info
->trans_lock
);
1818 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
,
1819 struct btrfs_root
*root
)
1821 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1822 struct btrfs_transaction
*prev_trans
= NULL
;
1823 struct btrfs_inode
*btree_ino
= BTRFS_I(root
->fs_info
->btree_inode
);
1826 /* Stop the commit early if ->aborted is set */
1827 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1828 ret
= cur_trans
->aborted
;
1829 btrfs_end_transaction(trans
, root
);
1833 /* make a pass through all the delayed refs we have so far
1834 * any runnings procs may add more while we are here
1836 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1838 btrfs_end_transaction(trans
, root
);
1842 btrfs_trans_release_metadata(trans
, root
);
1843 trans
->block_rsv
= NULL
;
1844 if (trans
->qgroup_reserved
) {
1845 btrfs_qgroup_free(root
, trans
->qgroup_reserved
);
1846 trans
->qgroup_reserved
= 0;
1849 cur_trans
= trans
->transaction
;
1852 * set the flushing flag so procs in this transaction have to
1853 * start sending their work down.
1855 cur_trans
->delayed_refs
.flushing
= 1;
1858 if (!list_empty(&trans
->new_bgs
))
1859 btrfs_create_pending_block_groups(trans
, root
);
1861 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1863 btrfs_end_transaction(trans
, root
);
1867 if (!cur_trans
->dirty_bg_run
) {
1870 /* this mutex is also taken before trying to set
1871 * block groups readonly. We need to make sure
1872 * that nobody has set a block group readonly
1873 * after a extents from that block group have been
1874 * allocated for cache files. btrfs_set_block_group_ro
1875 * will wait for the transaction to commit if it
1876 * finds dirty_bg_run = 1
1878 * The dirty_bg_run flag is also used to make sure only
1879 * one process starts all the block group IO. It wouldn't
1880 * hurt to have more than one go through, but there's no
1881 * real advantage to it either.
1883 mutex_lock(&root
->fs_info
->ro_block_group_mutex
);
1884 if (!cur_trans
->dirty_bg_run
) {
1886 cur_trans
->dirty_bg_run
= 1;
1888 mutex_unlock(&root
->fs_info
->ro_block_group_mutex
);
1891 ret
= btrfs_start_dirty_block_groups(trans
, root
);
1894 btrfs_end_transaction(trans
, root
);
1898 spin_lock(&root
->fs_info
->trans_lock
);
1899 list_splice_init(&trans
->ordered
, &cur_trans
->pending_ordered
);
1900 if (cur_trans
->state
>= TRANS_STATE_COMMIT_START
) {
1901 spin_unlock(&root
->fs_info
->trans_lock
);
1902 atomic_inc(&cur_trans
->use_count
);
1903 ret
= btrfs_end_transaction(trans
, root
);
1905 wait_for_commit(root
, cur_trans
);
1907 if (unlikely(cur_trans
->aborted
))
1908 ret
= cur_trans
->aborted
;
1910 btrfs_put_transaction(cur_trans
);
1915 cur_trans
->state
= TRANS_STATE_COMMIT_START
;
1916 wake_up(&root
->fs_info
->transaction_blocked_wait
);
1918 if (cur_trans
->list
.prev
!= &root
->fs_info
->trans_list
) {
1919 prev_trans
= list_entry(cur_trans
->list
.prev
,
1920 struct btrfs_transaction
, list
);
1921 if (prev_trans
->state
!= TRANS_STATE_COMPLETED
) {
1922 atomic_inc(&prev_trans
->use_count
);
1923 spin_unlock(&root
->fs_info
->trans_lock
);
1925 wait_for_commit(root
, prev_trans
);
1926 ret
= prev_trans
->aborted
;
1928 btrfs_put_transaction(prev_trans
);
1930 goto cleanup_transaction
;
1932 spin_unlock(&root
->fs_info
->trans_lock
);
1935 spin_unlock(&root
->fs_info
->trans_lock
);
1938 extwriter_counter_dec(cur_trans
, trans
->type
);
1940 ret
= btrfs_start_delalloc_flush(root
->fs_info
);
1942 goto cleanup_transaction
;
1944 ret
= btrfs_run_delayed_items(trans
, root
);
1946 goto cleanup_transaction
;
1948 wait_event(cur_trans
->writer_wait
,
1949 extwriter_counter_read(cur_trans
) == 0);
1951 /* some pending stuffs might be added after the previous flush. */
1952 ret
= btrfs_run_delayed_items(trans
, root
);
1954 goto cleanup_transaction
;
1956 btrfs_wait_delalloc_flush(root
->fs_info
);
1958 btrfs_wait_pending_ordered(cur_trans
, root
->fs_info
);
1960 btrfs_scrub_pause(root
);
1962 * Ok now we need to make sure to block out any other joins while we
1963 * commit the transaction. We could have started a join before setting
1964 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
1966 spin_lock(&root
->fs_info
->trans_lock
);
1967 cur_trans
->state
= TRANS_STATE_COMMIT_DOING
;
1968 spin_unlock(&root
->fs_info
->trans_lock
);
1969 wait_event(cur_trans
->writer_wait
,
1970 atomic_read(&cur_trans
->num_writers
) == 1);
1972 /* ->aborted might be set after the previous check, so check it */
1973 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1974 ret
= cur_trans
->aborted
;
1975 goto scrub_continue
;
1978 * the reloc mutex makes sure that we stop
1979 * the balancing code from coming in and moving
1980 * extents around in the middle of the commit
1982 mutex_lock(&root
->fs_info
->reloc_mutex
);
1985 * We needn't worry about the delayed items because we will
1986 * deal with them in create_pending_snapshot(), which is the
1987 * core function of the snapshot creation.
1989 ret
= create_pending_snapshots(trans
, root
->fs_info
);
1991 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1992 goto scrub_continue
;
1996 * We insert the dir indexes of the snapshots and update the inode
1997 * of the snapshots' parents after the snapshot creation, so there
1998 * are some delayed items which are not dealt with. Now deal with
2001 * We needn't worry that this operation will corrupt the snapshots,
2002 * because all the tree which are snapshoted will be forced to COW
2003 * the nodes and leaves.
2005 ret
= btrfs_run_delayed_items(trans
, root
);
2007 mutex_unlock(&root
->fs_info
->reloc_mutex
);
2008 goto scrub_continue
;
2011 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
2013 mutex_unlock(&root
->fs_info
->reloc_mutex
);
2014 goto scrub_continue
;
2017 /* Reocrd old roots for later qgroup accounting */
2018 ret
= btrfs_qgroup_prepare_account_extents(trans
, root
->fs_info
);
2020 mutex_unlock(&root
->fs_info
->reloc_mutex
);
2021 goto scrub_continue
;
2025 * make sure none of the code above managed to slip in a
2028 btrfs_assert_delayed_root_empty(root
);
2030 WARN_ON(cur_trans
!= trans
->transaction
);
2032 /* btrfs_commit_tree_roots is responsible for getting the
2033 * various roots consistent with each other. Every pointer
2034 * in the tree of tree roots has to point to the most up to date
2035 * root for every subvolume and other tree. So, we have to keep
2036 * the tree logging code from jumping in and changing any
2039 * At this point in the commit, there can't be any tree-log
2040 * writers, but a little lower down we drop the trans mutex
2041 * and let new people in. By holding the tree_log_mutex
2042 * from now until after the super is written, we avoid races
2043 * with the tree-log code.
2045 mutex_lock(&root
->fs_info
->tree_log_mutex
);
2047 ret
= commit_fs_roots(trans
, root
);
2049 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
2050 mutex_unlock(&root
->fs_info
->reloc_mutex
);
2051 goto scrub_continue
;
2055 * Since the transaction is done, we can apply the pending changes
2056 * before the next transaction.
2058 btrfs_apply_pending_changes(root
->fs_info
);
2060 /* commit_fs_roots gets rid of all the tree log roots, it is now
2061 * safe to free the root of tree log roots
2063 btrfs_free_log_root_tree(trans
, root
->fs_info
);
2066 * Since fs roots are all committed, we can get a quite accurate
2067 * new_roots. So let's do quota accounting.
2069 ret
= btrfs_qgroup_account_extents(trans
, root
->fs_info
);
2071 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
2072 mutex_unlock(&root
->fs_info
->reloc_mutex
);
2073 goto scrub_continue
;
2076 ret
= commit_cowonly_roots(trans
, root
);
2078 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
2079 mutex_unlock(&root
->fs_info
->reloc_mutex
);
2080 goto scrub_continue
;
2084 * The tasks which save the space cache and inode cache may also
2085 * update ->aborted, check it.
2087 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
2088 ret
= cur_trans
->aborted
;
2089 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
2090 mutex_unlock(&root
->fs_info
->reloc_mutex
);
2091 goto scrub_continue
;
2094 btrfs_prepare_extent_commit(trans
, root
);
2096 cur_trans
= root
->fs_info
->running_transaction
;
2098 btrfs_set_root_node(&root
->fs_info
->tree_root
->root_item
,
2099 root
->fs_info
->tree_root
->node
);
2100 list_add_tail(&root
->fs_info
->tree_root
->dirty_list
,
2101 &cur_trans
->switch_commits
);
2103 btrfs_set_root_node(&root
->fs_info
->chunk_root
->root_item
,
2104 root
->fs_info
->chunk_root
->node
);
2105 list_add_tail(&root
->fs_info
->chunk_root
->dirty_list
,
2106 &cur_trans
->switch_commits
);
2108 switch_commit_roots(cur_trans
, root
->fs_info
);
2110 assert_qgroups_uptodate(trans
);
2111 ASSERT(list_empty(&cur_trans
->dirty_bgs
));
2112 ASSERT(list_empty(&cur_trans
->io_bgs
));
2113 update_super_roots(root
);
2115 btrfs_set_super_log_root(root
->fs_info
->super_copy
, 0);
2116 btrfs_set_super_log_root_level(root
->fs_info
->super_copy
, 0);
2117 memcpy(root
->fs_info
->super_for_commit
, root
->fs_info
->super_copy
,
2118 sizeof(*root
->fs_info
->super_copy
));
2120 btrfs_update_commit_device_size(root
->fs_info
);
2121 btrfs_update_commit_device_bytes_used(root
, cur_trans
);
2123 clear_bit(BTRFS_INODE_BTREE_LOG1_ERR
, &btree_ino
->runtime_flags
);
2124 clear_bit(BTRFS_INODE_BTREE_LOG2_ERR
, &btree_ino
->runtime_flags
);
2126 btrfs_trans_release_chunk_metadata(trans
);
2128 spin_lock(&root
->fs_info
->trans_lock
);
2129 cur_trans
->state
= TRANS_STATE_UNBLOCKED
;
2130 root
->fs_info
->running_transaction
= NULL
;
2131 spin_unlock(&root
->fs_info
->trans_lock
);
2132 mutex_unlock(&root
->fs_info
->reloc_mutex
);
2134 wake_up(&root
->fs_info
->transaction_wait
);
2136 ret
= btrfs_write_and_wait_transaction(trans
, root
);
2138 btrfs_error(root
->fs_info
, ret
,
2139 "Error while writing out transaction");
2140 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
2141 goto scrub_continue
;
2144 ret
= write_ctree_super(trans
, root
, 0);
2146 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
2147 goto scrub_continue
;
2151 * the super is written, we can safely allow the tree-loggers
2152 * to go about their business
2154 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
2156 btrfs_finish_extent_commit(trans
, root
);
2158 if (cur_trans
->have_free_bgs
)
2159 btrfs_clear_space_info_full(root
->fs_info
);
2161 root
->fs_info
->last_trans_committed
= cur_trans
->transid
;
2163 * We needn't acquire the lock here because there is no other task
2164 * which can change it.
2166 cur_trans
->state
= TRANS_STATE_COMPLETED
;
2167 wake_up(&cur_trans
->commit_wait
);
2169 spin_lock(&root
->fs_info
->trans_lock
);
2170 list_del_init(&cur_trans
->list
);
2171 spin_unlock(&root
->fs_info
->trans_lock
);
2173 btrfs_put_transaction(cur_trans
);
2174 btrfs_put_transaction(cur_trans
);
2176 if (trans
->type
& __TRANS_FREEZABLE
)
2177 sb_end_intwrite(root
->fs_info
->sb
);
2179 trace_btrfs_transaction_commit(root
);
2181 btrfs_scrub_continue(root
);
2183 if (current
->journal_info
== trans
)
2184 current
->journal_info
= NULL
;
2186 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
2188 if (current
!= root
->fs_info
->transaction_kthread
&&
2189 current
!= root
->fs_info
->cleaner_kthread
)
2190 btrfs_run_delayed_iputs(root
);
2195 btrfs_scrub_continue(root
);
2196 cleanup_transaction
:
2197 btrfs_trans_release_metadata(trans
, root
);
2198 btrfs_trans_release_chunk_metadata(trans
);
2199 trans
->block_rsv
= NULL
;
2200 if (trans
->qgroup_reserved
) {
2201 btrfs_qgroup_free(root
, trans
->qgroup_reserved
);
2202 trans
->qgroup_reserved
= 0;
2204 btrfs_warn(root
->fs_info
, "Skipping commit of aborted transaction.");
2205 if (current
->journal_info
== trans
)
2206 current
->journal_info
= NULL
;
2207 cleanup_transaction(trans
, root
, ret
);
2213 * return < 0 if error
2214 * 0 if there are no more dead_roots at the time of call
2215 * 1 there are more to be processed, call me again
2217 * The return value indicates there are certainly more snapshots to delete, but
2218 * if there comes a new one during processing, it may return 0. We don't mind,
2219 * because btrfs_commit_super will poke cleaner thread and it will process it a
2220 * few seconds later.
2222 int btrfs_clean_one_deleted_snapshot(struct btrfs_root
*root
)
2225 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2227 spin_lock(&fs_info
->trans_lock
);
2228 if (list_empty(&fs_info
->dead_roots
)) {
2229 spin_unlock(&fs_info
->trans_lock
);
2232 root
= list_first_entry(&fs_info
->dead_roots
,
2233 struct btrfs_root
, root_list
);
2234 list_del_init(&root
->root_list
);
2235 spin_unlock(&fs_info
->trans_lock
);
2237 pr_debug("BTRFS: cleaner removing %llu\n", root
->objectid
);
2239 btrfs_kill_all_delayed_nodes(root
);
2241 if (btrfs_header_backref_rev(root
->node
) <
2242 BTRFS_MIXED_BACKREF_REV
)
2243 ret
= btrfs_drop_snapshot(root
, NULL
, 0, 0);
2245 ret
= btrfs_drop_snapshot(root
, NULL
, 1, 0);
2247 return (ret
< 0) ? 0 : 1;
2250 void btrfs_apply_pending_changes(struct btrfs_fs_info
*fs_info
)
2255 prev
= xchg(&fs_info
->pending_changes
, 0);
2259 bit
= 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE
;
2261 btrfs_set_opt(fs_info
->mount_opt
, INODE_MAP_CACHE
);
2264 bit
= 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE
;
2266 btrfs_clear_opt(fs_info
->mount_opt
, INODE_MAP_CACHE
);
2269 bit
= 1 << BTRFS_PENDING_COMMIT
;
2271 btrfs_debug(fs_info
, "pending commit done");
2276 "unknown pending changes left 0x%lx, ignoring", prev
);