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
);
86 * Do a single barrier for the waitqueue_active check here, the state
87 * of the waitqueue should not change once clear_btree_io_tree is
91 while (!RB_EMPTY_ROOT(&tree
->state
)) {
93 struct extent_state
*state
;
95 node
= rb_first(&tree
->state
);
96 state
= rb_entry(node
, struct extent_state
, rb_node
);
97 rb_erase(&state
->rb_node
, &tree
->state
);
98 RB_CLEAR_NODE(&state
->rb_node
);
100 * btree io trees aren't supposed to have tasks waiting for
101 * changes in the flags of extent states ever.
103 ASSERT(!waitqueue_active(&state
->wq
));
104 free_extent_state(state
);
106 cond_resched_lock(&tree
->lock
);
108 spin_unlock(&tree
->lock
);
111 static noinline
void switch_commit_roots(struct btrfs_transaction
*trans
,
112 struct btrfs_fs_info
*fs_info
)
114 struct btrfs_root
*root
, *tmp
;
116 down_write(&fs_info
->commit_root_sem
);
117 list_for_each_entry_safe(root
, tmp
, &trans
->switch_commits
,
119 list_del_init(&root
->dirty_list
);
120 free_extent_buffer(root
->commit_root
);
121 root
->commit_root
= btrfs_root_node(root
);
122 if (is_fstree(root
->objectid
))
123 btrfs_unpin_free_ino(root
);
124 clear_btree_io_tree(&root
->dirty_log_pages
);
127 /* We can free old roots now. */
128 spin_lock(&trans
->dropped_roots_lock
);
129 while (!list_empty(&trans
->dropped_roots
)) {
130 root
= list_first_entry(&trans
->dropped_roots
,
131 struct btrfs_root
, root_list
);
132 list_del_init(&root
->root_list
);
133 spin_unlock(&trans
->dropped_roots_lock
);
134 btrfs_drop_and_free_fs_root(fs_info
, root
);
135 spin_lock(&trans
->dropped_roots_lock
);
137 spin_unlock(&trans
->dropped_roots_lock
);
138 up_write(&fs_info
->commit_root_sem
);
141 static inline void extwriter_counter_inc(struct btrfs_transaction
*trans
,
144 if (type
& TRANS_EXTWRITERS
)
145 atomic_inc(&trans
->num_extwriters
);
148 static inline void extwriter_counter_dec(struct btrfs_transaction
*trans
,
151 if (type
& TRANS_EXTWRITERS
)
152 atomic_dec(&trans
->num_extwriters
);
155 static inline void extwriter_counter_init(struct btrfs_transaction
*trans
,
158 atomic_set(&trans
->num_extwriters
, ((type
& TRANS_EXTWRITERS
) ? 1 : 0));
161 static inline int extwriter_counter_read(struct btrfs_transaction
*trans
)
163 return atomic_read(&trans
->num_extwriters
);
167 * either allocate a new transaction or hop into the existing one
169 static noinline
int join_transaction(struct btrfs_root
*root
, unsigned int type
)
171 struct btrfs_transaction
*cur_trans
;
172 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
174 spin_lock(&fs_info
->trans_lock
);
176 /* The file system has been taken offline. No new transactions. */
177 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
178 spin_unlock(&fs_info
->trans_lock
);
182 cur_trans
= fs_info
->running_transaction
;
184 if (cur_trans
->aborted
) {
185 spin_unlock(&fs_info
->trans_lock
);
186 return cur_trans
->aborted
;
188 if (btrfs_blocked_trans_types
[cur_trans
->state
] & type
) {
189 spin_unlock(&fs_info
->trans_lock
);
192 atomic_inc(&cur_trans
->use_count
);
193 atomic_inc(&cur_trans
->num_writers
);
194 extwriter_counter_inc(cur_trans
, type
);
195 spin_unlock(&fs_info
->trans_lock
);
198 spin_unlock(&fs_info
->trans_lock
);
201 * If we are ATTACH, we just want to catch the current transaction,
202 * and commit it. If there is no transaction, just return ENOENT.
204 if (type
== TRANS_ATTACH
)
208 * JOIN_NOLOCK only happens during the transaction commit, so
209 * it is impossible that ->running_transaction is NULL
211 BUG_ON(type
== TRANS_JOIN_NOLOCK
);
213 cur_trans
= kmem_cache_alloc(btrfs_transaction_cachep
, GFP_NOFS
);
217 spin_lock(&fs_info
->trans_lock
);
218 if (fs_info
->running_transaction
) {
220 * someone started a transaction after we unlocked. Make sure
221 * to redo the checks above
223 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
225 } else if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
226 spin_unlock(&fs_info
->trans_lock
);
227 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
231 atomic_set(&cur_trans
->num_writers
, 1);
232 extwriter_counter_init(cur_trans
, type
);
233 init_waitqueue_head(&cur_trans
->writer_wait
);
234 init_waitqueue_head(&cur_trans
->commit_wait
);
235 cur_trans
->state
= TRANS_STATE_RUNNING
;
237 * One for this trans handle, one so it will live on until we
238 * commit the transaction.
240 atomic_set(&cur_trans
->use_count
, 2);
241 cur_trans
->have_free_bgs
= 0;
242 cur_trans
->start_time
= get_seconds();
243 cur_trans
->dirty_bg_run
= 0;
245 cur_trans
->delayed_refs
.href_root
= RB_ROOT
;
246 cur_trans
->delayed_refs
.dirty_extent_root
= RB_ROOT
;
247 atomic_set(&cur_trans
->delayed_refs
.num_entries
, 0);
248 cur_trans
->delayed_refs
.num_heads_ready
= 0;
249 cur_trans
->delayed_refs
.pending_csums
= 0;
250 cur_trans
->delayed_refs
.num_heads
= 0;
251 cur_trans
->delayed_refs
.flushing
= 0;
252 cur_trans
->delayed_refs
.run_delayed_start
= 0;
253 cur_trans
->delayed_refs
.qgroup_to_skip
= 0;
256 * although the tree mod log is per file system and not per transaction,
257 * the log must never go across transaction boundaries.
260 if (!list_empty(&fs_info
->tree_mod_seq_list
))
261 WARN(1, KERN_ERR
"BTRFS: tree_mod_seq_list not empty when "
262 "creating a fresh transaction\n");
263 if (!RB_EMPTY_ROOT(&fs_info
->tree_mod_log
))
264 WARN(1, KERN_ERR
"BTRFS: tree_mod_log rb tree not empty when "
265 "creating a fresh transaction\n");
266 atomic64_set(&fs_info
->tree_mod_seq
, 0);
268 spin_lock_init(&cur_trans
->delayed_refs
.lock
);
270 INIT_LIST_HEAD(&cur_trans
->pending_snapshots
);
271 INIT_LIST_HEAD(&cur_trans
->pending_chunks
);
272 INIT_LIST_HEAD(&cur_trans
->switch_commits
);
273 INIT_LIST_HEAD(&cur_trans
->pending_ordered
);
274 INIT_LIST_HEAD(&cur_trans
->dirty_bgs
);
275 INIT_LIST_HEAD(&cur_trans
->io_bgs
);
276 INIT_LIST_HEAD(&cur_trans
->dropped_roots
);
277 mutex_init(&cur_trans
->cache_write_mutex
);
278 cur_trans
->num_dirty_bgs
= 0;
279 spin_lock_init(&cur_trans
->dirty_bgs_lock
);
280 INIT_LIST_HEAD(&cur_trans
->deleted_bgs
);
281 spin_lock_init(&cur_trans
->deleted_bgs_lock
);
282 spin_lock_init(&cur_trans
->dropped_roots_lock
);
283 list_add_tail(&cur_trans
->list
, &fs_info
->trans_list
);
284 extent_io_tree_init(&cur_trans
->dirty_pages
,
285 fs_info
->btree_inode
->i_mapping
);
286 fs_info
->generation
++;
287 cur_trans
->transid
= fs_info
->generation
;
288 fs_info
->running_transaction
= cur_trans
;
289 cur_trans
->aborted
= 0;
290 spin_unlock(&fs_info
->trans_lock
);
296 * this does all the record keeping required to make sure that a reference
297 * counted root is properly recorded in a given transaction. This is required
298 * to make sure the old root from before we joined the transaction is deleted
299 * when the transaction commits
301 static int record_root_in_trans(struct btrfs_trans_handle
*trans
,
302 struct btrfs_root
*root
)
304 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) &&
305 root
->last_trans
< trans
->transid
) {
306 WARN_ON(root
== root
->fs_info
->extent_root
);
307 WARN_ON(root
->commit_root
!= root
->node
);
310 * see below for IN_TRANS_SETUP usage rules
311 * we have the reloc mutex held now, so there
312 * is only one writer in this function
314 set_bit(BTRFS_ROOT_IN_TRANS_SETUP
, &root
->state
);
316 /* make sure readers find IN_TRANS_SETUP before
317 * they find our root->last_trans update
321 spin_lock(&root
->fs_info
->fs_roots_radix_lock
);
322 if (root
->last_trans
== trans
->transid
) {
323 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
326 radix_tree_tag_set(&root
->fs_info
->fs_roots_radix
,
327 (unsigned long)root
->root_key
.objectid
,
328 BTRFS_ROOT_TRANS_TAG
);
329 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
330 root
->last_trans
= trans
->transid
;
332 /* this is pretty tricky. We don't want to
333 * take the relocation lock in btrfs_record_root_in_trans
334 * unless we're really doing the first setup for this root in
337 * Normally we'd use root->last_trans as a flag to decide
338 * if we want to take the expensive mutex.
340 * But, we have to set root->last_trans before we
341 * init the relocation root, otherwise, we trip over warnings
342 * in ctree.c. The solution used here is to flag ourselves
343 * with root IN_TRANS_SETUP. When this is 1, we're still
344 * fixing up the reloc trees and everyone must wait.
346 * When this is zero, they can trust root->last_trans and fly
347 * through btrfs_record_root_in_trans without having to take the
348 * lock. smp_wmb() makes sure that all the writes above are
349 * done before we pop in the zero below
351 btrfs_init_reloc_root(trans
, root
);
352 smp_mb__before_atomic();
353 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP
, &root
->state
);
359 void btrfs_add_dropped_root(struct btrfs_trans_handle
*trans
,
360 struct btrfs_root
*root
)
362 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
364 /* Add ourselves to the transaction dropped list */
365 spin_lock(&cur_trans
->dropped_roots_lock
);
366 list_add_tail(&root
->root_list
, &cur_trans
->dropped_roots
);
367 spin_unlock(&cur_trans
->dropped_roots_lock
);
369 /* Make sure we don't try to update the root at commit time */
370 spin_lock(&root
->fs_info
->fs_roots_radix_lock
);
371 radix_tree_tag_clear(&root
->fs_info
->fs_roots_radix
,
372 (unsigned long)root
->root_key
.objectid
,
373 BTRFS_ROOT_TRANS_TAG
);
374 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
377 int btrfs_record_root_in_trans(struct btrfs_trans_handle
*trans
,
378 struct btrfs_root
*root
)
380 if (!test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
384 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
388 if (root
->last_trans
== trans
->transid
&&
389 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP
, &root
->state
))
392 mutex_lock(&root
->fs_info
->reloc_mutex
);
393 record_root_in_trans(trans
, root
);
394 mutex_unlock(&root
->fs_info
->reloc_mutex
);
399 static inline int is_transaction_blocked(struct btrfs_transaction
*trans
)
401 return (trans
->state
>= TRANS_STATE_BLOCKED
&&
402 trans
->state
< TRANS_STATE_UNBLOCKED
&&
406 /* wait for commit against the current transaction to become unblocked
407 * when this is done, it is safe to start a new transaction, but the current
408 * transaction might not be fully on disk.
410 static void wait_current_trans(struct btrfs_root
*root
)
412 struct btrfs_transaction
*cur_trans
;
414 spin_lock(&root
->fs_info
->trans_lock
);
415 cur_trans
= root
->fs_info
->running_transaction
;
416 if (cur_trans
&& is_transaction_blocked(cur_trans
)) {
417 atomic_inc(&cur_trans
->use_count
);
418 spin_unlock(&root
->fs_info
->trans_lock
);
420 wait_event(root
->fs_info
->transaction_wait
,
421 cur_trans
->state
>= TRANS_STATE_UNBLOCKED
||
423 btrfs_put_transaction(cur_trans
);
425 spin_unlock(&root
->fs_info
->trans_lock
);
429 static int may_wait_transaction(struct btrfs_root
*root
, int type
)
431 if (root
->fs_info
->log_root_recovering
)
434 if (type
== TRANS_USERSPACE
)
437 if (type
== TRANS_START
&&
438 !atomic_read(&root
->fs_info
->open_ioctl_trans
))
444 static inline bool need_reserve_reloc_root(struct btrfs_root
*root
)
446 if (!root
->fs_info
->reloc_ctl
||
447 !test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
448 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
455 static struct btrfs_trans_handle
*
456 start_transaction(struct btrfs_root
*root
, u64 num_items
, unsigned int type
,
457 enum btrfs_reserve_flush_enum flush
)
459 struct btrfs_trans_handle
*h
;
460 struct btrfs_transaction
*cur_trans
;
462 u64 qgroup_reserved
= 0;
463 bool reloc_reserved
= false;
466 /* Send isn't supposed to start transactions. */
467 ASSERT(current
->journal_info
!= BTRFS_SEND_TRANS_STUB
);
469 if (test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
470 return ERR_PTR(-EROFS
);
472 if (current
->journal_info
) {
473 WARN_ON(type
& TRANS_EXTWRITERS
);
474 h
= current
->journal_info
;
476 WARN_ON(h
->use_count
> 2);
477 h
->orig_rsv
= h
->block_rsv
;
483 * Do the reservation before we join the transaction so we can do all
484 * the appropriate flushing if need be.
486 if (num_items
> 0 && root
!= root
->fs_info
->chunk_root
) {
487 if (root
->fs_info
->quota_enabled
&&
488 is_fstree(root
->root_key
.objectid
)) {
489 qgroup_reserved
= num_items
* root
->nodesize
;
490 ret
= btrfs_qgroup_reserve(root
, qgroup_reserved
);
495 num_bytes
= btrfs_calc_trans_metadata_size(root
, num_items
);
497 * Do the reservation for the relocation root creation
499 if (need_reserve_reloc_root(root
)) {
500 num_bytes
+= root
->nodesize
;
501 reloc_reserved
= true;
504 ret
= btrfs_block_rsv_add(root
,
505 &root
->fs_info
->trans_block_rsv
,
511 h
= kmem_cache_alloc(btrfs_trans_handle_cachep
, GFP_NOFS
);
518 * If we are JOIN_NOLOCK we're already committing a transaction and
519 * waiting on this guy, so we don't need to do the sb_start_intwrite
520 * because we're already holding a ref. We need this because we could
521 * have raced in and did an fsync() on a file which can kick a commit
522 * and then we deadlock with somebody doing a freeze.
524 * If we are ATTACH, it means we just want to catch the current
525 * transaction and commit it, so we needn't do sb_start_intwrite().
527 if (type
& __TRANS_FREEZABLE
)
528 sb_start_intwrite(root
->fs_info
->sb
);
530 if (may_wait_transaction(root
, type
))
531 wait_current_trans(root
);
534 ret
= join_transaction(root
, type
);
536 wait_current_trans(root
);
537 if (unlikely(type
== TRANS_ATTACH
))
540 } while (ret
== -EBUSY
);
543 /* We must get the transaction if we are JOIN_NOLOCK. */
544 BUG_ON(type
== TRANS_JOIN_NOLOCK
);
548 cur_trans
= root
->fs_info
->running_transaction
;
550 h
->transid
= cur_trans
->transid
;
551 h
->transaction
= cur_trans
;
553 h
->bytes_reserved
= 0;
554 h
->chunk_bytes_reserved
= 0;
556 h
->delayed_ref_updates
= 0;
562 h
->qgroup_reserved
= 0;
563 h
->delayed_ref_elem
.seq
= 0;
565 h
->allocating_chunk
= false;
566 h
->reloc_reserved
= false;
568 INIT_LIST_HEAD(&h
->qgroup_ref_list
);
569 INIT_LIST_HEAD(&h
->new_bgs
);
570 INIT_LIST_HEAD(&h
->ordered
);
573 if (cur_trans
->state
>= TRANS_STATE_BLOCKED
&&
574 may_wait_transaction(root
, type
)) {
575 current
->journal_info
= h
;
576 btrfs_commit_transaction(h
, root
);
581 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
582 h
->transid
, num_bytes
, 1);
583 h
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
584 h
->bytes_reserved
= num_bytes
;
585 h
->reloc_reserved
= reloc_reserved
;
587 h
->qgroup_reserved
= qgroup_reserved
;
590 btrfs_record_root_in_trans(h
, root
);
592 if (!current
->journal_info
&& type
!= TRANS_USERSPACE
)
593 current
->journal_info
= h
;
597 if (type
& __TRANS_FREEZABLE
)
598 sb_end_intwrite(root
->fs_info
->sb
);
599 kmem_cache_free(btrfs_trans_handle_cachep
, h
);
602 btrfs_block_rsv_release(root
, &root
->fs_info
->trans_block_rsv
,
606 btrfs_qgroup_free(root
, qgroup_reserved
);
610 struct btrfs_trans_handle
*btrfs_start_transaction(struct btrfs_root
*root
,
613 return start_transaction(root
, num_items
, TRANS_START
,
614 BTRFS_RESERVE_FLUSH_ALL
);
617 struct btrfs_trans_handle
*btrfs_start_transaction_lflush(
618 struct btrfs_root
*root
, int num_items
)
620 return start_transaction(root
, num_items
, TRANS_START
,
621 BTRFS_RESERVE_FLUSH_LIMIT
);
624 struct btrfs_trans_handle
*btrfs_join_transaction(struct btrfs_root
*root
)
626 return start_transaction(root
, 0, TRANS_JOIN
, 0);
629 struct btrfs_trans_handle
*btrfs_join_transaction_nolock(struct btrfs_root
*root
)
631 return start_transaction(root
, 0, TRANS_JOIN_NOLOCK
, 0);
634 struct btrfs_trans_handle
*btrfs_start_ioctl_transaction(struct btrfs_root
*root
)
636 return start_transaction(root
, 0, TRANS_USERSPACE
, 0);
640 * btrfs_attach_transaction() - catch the running transaction
642 * It is used when we want to commit the current the transaction, but
643 * don't want to start a new one.
645 * Note: If this function return -ENOENT, it just means there is no
646 * running transaction. But it is possible that the inactive transaction
647 * is still in the memory, not fully on disk. If you hope there is no
648 * inactive transaction in the fs when -ENOENT is returned, you should
650 * btrfs_attach_transaction_barrier()
652 struct btrfs_trans_handle
*btrfs_attach_transaction(struct btrfs_root
*root
)
654 return start_transaction(root
, 0, TRANS_ATTACH
, 0);
658 * btrfs_attach_transaction_barrier() - catch the running transaction
660 * It is similar to the above function, the differentia is this one
661 * will wait for all the inactive transactions until they fully
664 struct btrfs_trans_handle
*
665 btrfs_attach_transaction_barrier(struct btrfs_root
*root
)
667 struct btrfs_trans_handle
*trans
;
669 trans
= start_transaction(root
, 0, TRANS_ATTACH
, 0);
670 if (IS_ERR(trans
) && PTR_ERR(trans
) == -ENOENT
)
671 btrfs_wait_for_commit(root
, 0);
676 /* wait for a transaction commit to be fully complete */
677 static noinline
void wait_for_commit(struct btrfs_root
*root
,
678 struct btrfs_transaction
*commit
)
680 wait_event(commit
->commit_wait
, commit
->state
== TRANS_STATE_COMPLETED
);
683 int btrfs_wait_for_commit(struct btrfs_root
*root
, u64 transid
)
685 struct btrfs_transaction
*cur_trans
= NULL
, *t
;
689 if (transid
<= root
->fs_info
->last_trans_committed
)
692 /* find specified transaction */
693 spin_lock(&root
->fs_info
->trans_lock
);
694 list_for_each_entry(t
, &root
->fs_info
->trans_list
, list
) {
695 if (t
->transid
== transid
) {
697 atomic_inc(&cur_trans
->use_count
);
701 if (t
->transid
> transid
) {
706 spin_unlock(&root
->fs_info
->trans_lock
);
709 * The specified transaction doesn't exist, or we
710 * raced with btrfs_commit_transaction
713 if (transid
> root
->fs_info
->last_trans_committed
)
718 /* find newest transaction that is committing | committed */
719 spin_lock(&root
->fs_info
->trans_lock
);
720 list_for_each_entry_reverse(t
, &root
->fs_info
->trans_list
,
722 if (t
->state
>= TRANS_STATE_COMMIT_START
) {
723 if (t
->state
== TRANS_STATE_COMPLETED
)
726 atomic_inc(&cur_trans
->use_count
);
730 spin_unlock(&root
->fs_info
->trans_lock
);
732 goto out
; /* nothing committing|committed */
735 wait_for_commit(root
, cur_trans
);
736 btrfs_put_transaction(cur_trans
);
741 void btrfs_throttle(struct btrfs_root
*root
)
743 if (!atomic_read(&root
->fs_info
->open_ioctl_trans
))
744 wait_current_trans(root
);
747 static int should_end_transaction(struct btrfs_trans_handle
*trans
,
748 struct btrfs_root
*root
)
750 if (root
->fs_info
->global_block_rsv
.space_info
->full
&&
751 btrfs_check_space_for_delayed_refs(trans
, root
))
754 return !!btrfs_block_rsv_check(root
, &root
->fs_info
->global_block_rsv
, 5);
757 int btrfs_should_end_transaction(struct btrfs_trans_handle
*trans
,
758 struct btrfs_root
*root
)
760 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
765 if (cur_trans
->state
>= TRANS_STATE_BLOCKED
||
766 cur_trans
->delayed_refs
.flushing
)
769 updates
= trans
->delayed_ref_updates
;
770 trans
->delayed_ref_updates
= 0;
772 err
= btrfs_run_delayed_refs(trans
, root
, updates
* 2);
773 if (err
) /* Error code will also eval true */
777 return should_end_transaction(trans
, root
);
780 static int __btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
781 struct btrfs_root
*root
, int throttle
)
783 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
784 struct btrfs_fs_info
*info
= root
->fs_info
;
785 unsigned long cur
= trans
->delayed_ref_updates
;
786 int lock
= (trans
->type
!= TRANS_JOIN_NOLOCK
);
788 int must_run_delayed_refs
= 0;
790 if (trans
->use_count
> 1) {
792 trans
->block_rsv
= trans
->orig_rsv
;
796 btrfs_trans_release_metadata(trans
, root
);
797 trans
->block_rsv
= NULL
;
799 if (!list_empty(&trans
->new_bgs
))
800 btrfs_create_pending_block_groups(trans
, root
);
802 if (!list_empty(&trans
->ordered
)) {
803 spin_lock(&info
->trans_lock
);
804 list_splice_init(&trans
->ordered
, &cur_trans
->pending_ordered
);
805 spin_unlock(&info
->trans_lock
);
808 trans
->delayed_ref_updates
= 0;
810 must_run_delayed_refs
=
811 btrfs_should_throttle_delayed_refs(trans
, root
);
812 cur
= max_t(unsigned long, cur
, 32);
815 * don't make the caller wait if they are from a NOLOCK
816 * or ATTACH transaction, it will deadlock with commit
818 if (must_run_delayed_refs
== 1 &&
819 (trans
->type
& (__TRANS_JOIN_NOLOCK
| __TRANS_ATTACH
)))
820 must_run_delayed_refs
= 2;
823 if (trans
->qgroup_reserved
) {
825 * the same root has to be passed here between start_transaction
826 * and end_transaction. Subvolume quota depends on this.
828 btrfs_qgroup_free(trans
->root
, trans
->qgroup_reserved
);
829 trans
->qgroup_reserved
= 0;
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 btrfs_trans_release_chunk_metadata(trans
);
840 if (lock
&& !atomic_read(&root
->fs_info
->open_ioctl_trans
) &&
841 should_end_transaction(trans
, root
) &&
842 ACCESS_ONCE(cur_trans
->state
) == TRANS_STATE_RUNNING
) {
843 spin_lock(&info
->trans_lock
);
844 if (cur_trans
->state
== TRANS_STATE_RUNNING
)
845 cur_trans
->state
= TRANS_STATE_BLOCKED
;
846 spin_unlock(&info
->trans_lock
);
849 if (lock
&& ACCESS_ONCE(cur_trans
->state
) == TRANS_STATE_BLOCKED
) {
851 return btrfs_commit_transaction(trans
, root
);
853 wake_up_process(info
->transaction_kthread
);
856 if (trans
->type
& __TRANS_FREEZABLE
)
857 sb_end_intwrite(root
->fs_info
->sb
);
859 WARN_ON(cur_trans
!= info
->running_transaction
);
860 WARN_ON(atomic_read(&cur_trans
->num_writers
) < 1);
861 atomic_dec(&cur_trans
->num_writers
);
862 extwriter_counter_dec(cur_trans
, trans
->type
);
865 if (waitqueue_active(&cur_trans
->writer_wait
))
866 wake_up(&cur_trans
->writer_wait
);
867 btrfs_put_transaction(cur_trans
);
869 if (current
->journal_info
== trans
)
870 current
->journal_info
= NULL
;
873 btrfs_run_delayed_iputs(root
);
875 if (trans
->aborted
||
876 test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
)) {
877 wake_up_process(info
->transaction_kthread
);
880 assert_qgroups_uptodate(trans
);
882 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
883 if (must_run_delayed_refs
) {
884 btrfs_async_run_delayed_refs(root
, cur
,
885 must_run_delayed_refs
== 1);
890 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
891 struct btrfs_root
*root
)
893 return __btrfs_end_transaction(trans
, root
, 0);
896 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
,
897 struct btrfs_root
*root
)
899 return __btrfs_end_transaction(trans
, root
, 1);
903 * when btree blocks are allocated, they have some corresponding bits set for
904 * them in one of two extent_io trees. This is used to make sure all of
905 * those extents are sent to disk but does not wait on them
907 int btrfs_write_marked_extents(struct btrfs_root
*root
,
908 struct extent_io_tree
*dirty_pages
, int mark
)
912 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
913 struct extent_state
*cached_state
= NULL
;
917 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
918 mark
, &cached_state
)) {
919 bool wait_writeback
= false;
921 err
= convert_extent_bit(dirty_pages
, start
, end
,
923 mark
, &cached_state
, GFP_NOFS
);
925 * convert_extent_bit can return -ENOMEM, which is most of the
926 * time a temporary error. So when it happens, ignore the error
927 * and wait for writeback of this range to finish - because we
928 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
929 * to btrfs_wait_marked_extents() would not know that writeback
930 * for this range started and therefore wouldn't wait for it to
931 * finish - we don't want to commit a superblock that points to
932 * btree nodes/leafs for which writeback hasn't finished yet
933 * (and without errors).
934 * We cleanup any entries left in the io tree when committing
935 * the transaction (through clear_btree_io_tree()).
937 if (err
== -ENOMEM
) {
939 wait_writeback
= true;
942 err
= filemap_fdatawrite_range(mapping
, start
, end
);
945 else if (wait_writeback
)
946 werr
= filemap_fdatawait_range(mapping
, start
, end
);
947 free_extent_state(cached_state
);
956 * when btree blocks are allocated, they have some corresponding bits set for
957 * them in one of two extent_io trees. This is used to make sure all of
958 * those extents are on disk for transaction or log commit. We wait
959 * on all the pages and clear them from the dirty pages state tree
961 int btrfs_wait_marked_extents(struct btrfs_root
*root
,
962 struct extent_io_tree
*dirty_pages
, int mark
)
966 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
967 struct extent_state
*cached_state
= NULL
;
970 struct btrfs_inode
*btree_ino
= BTRFS_I(root
->fs_info
->btree_inode
);
973 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
974 EXTENT_NEED_WAIT
, &cached_state
)) {
976 * Ignore -ENOMEM errors returned by clear_extent_bit().
977 * When committing the transaction, we'll remove any entries
978 * left in the io tree. For a log commit, we don't remove them
979 * after committing the log because the tree can be accessed
980 * concurrently - we do it only at transaction commit time when
981 * it's safe to do it (through clear_btree_io_tree()).
983 err
= clear_extent_bit(dirty_pages
, start
, end
,
985 0, 0, &cached_state
, GFP_NOFS
);
989 err
= filemap_fdatawait_range(mapping
, start
, end
);
992 free_extent_state(cached_state
);
1000 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
1001 if ((mark
& EXTENT_DIRTY
) &&
1002 test_and_clear_bit(BTRFS_INODE_BTREE_LOG1_ERR
,
1003 &btree_ino
->runtime_flags
))
1006 if ((mark
& EXTENT_NEW
) &&
1007 test_and_clear_bit(BTRFS_INODE_BTREE_LOG2_ERR
,
1008 &btree_ino
->runtime_flags
))
1011 if (test_and_clear_bit(BTRFS_INODE_BTREE_ERR
,
1012 &btree_ino
->runtime_flags
))
1016 if (errors
&& !werr
)
1023 * when btree blocks are allocated, they have some corresponding bits set for
1024 * them in one of two extent_io trees. This is used to make sure all of
1025 * those extents are on disk for transaction or log commit
1027 static int btrfs_write_and_wait_marked_extents(struct btrfs_root
*root
,
1028 struct extent_io_tree
*dirty_pages
, int mark
)
1032 struct blk_plug plug
;
1034 blk_start_plug(&plug
);
1035 ret
= btrfs_write_marked_extents(root
, dirty_pages
, mark
);
1036 blk_finish_plug(&plug
);
1037 ret2
= btrfs_wait_marked_extents(root
, dirty_pages
, mark
);
1046 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
,
1047 struct btrfs_root
*root
)
1051 ret
= btrfs_write_and_wait_marked_extents(root
,
1052 &trans
->transaction
->dirty_pages
,
1054 clear_btree_io_tree(&trans
->transaction
->dirty_pages
);
1060 * this is used to update the root pointer in the tree of tree roots.
1062 * But, in the case of the extent allocation tree, updating the root
1063 * pointer may allocate blocks which may change the root of the extent
1066 * So, this loops and repeats and makes sure the cowonly root didn't
1067 * change while the root pointer was being updated in the metadata.
1069 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
1070 struct btrfs_root
*root
)
1073 u64 old_root_bytenr
;
1075 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
1077 old_root_used
= btrfs_root_used(&root
->root_item
);
1080 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
1081 if (old_root_bytenr
== root
->node
->start
&&
1082 old_root_used
== btrfs_root_used(&root
->root_item
))
1085 btrfs_set_root_node(&root
->root_item
, root
->node
);
1086 ret
= btrfs_update_root(trans
, tree_root
,
1092 old_root_used
= btrfs_root_used(&root
->root_item
);
1099 * update all the cowonly tree roots on disk
1101 * The error handling in this function may not be obvious. Any of the
1102 * failures will cause the file system to go offline. We still need
1103 * to clean up the delayed refs.
1105 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
,
1106 struct btrfs_root
*root
)
1108 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1109 struct list_head
*dirty_bgs
= &trans
->transaction
->dirty_bgs
;
1110 struct list_head
*io_bgs
= &trans
->transaction
->io_bgs
;
1111 struct list_head
*next
;
1112 struct extent_buffer
*eb
;
1115 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
1116 ret
= btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
,
1118 btrfs_tree_unlock(eb
);
1119 free_extent_buffer(eb
);
1124 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1128 ret
= btrfs_run_dev_stats(trans
, root
->fs_info
);
1131 ret
= btrfs_run_dev_replace(trans
, root
->fs_info
);
1134 ret
= btrfs_run_qgroups(trans
, root
->fs_info
);
1138 ret
= btrfs_setup_space_cache(trans
, root
);
1142 /* run_qgroups might have added some more refs */
1143 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1147 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
1148 next
= fs_info
->dirty_cowonly_roots
.next
;
1149 list_del_init(next
);
1150 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
1151 clear_bit(BTRFS_ROOT_DIRTY
, &root
->state
);
1153 if (root
!= fs_info
->extent_root
)
1154 list_add_tail(&root
->dirty_list
,
1155 &trans
->transaction
->switch_commits
);
1156 ret
= update_cowonly_root(trans
, root
);
1159 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1164 while (!list_empty(dirty_bgs
) || !list_empty(io_bgs
)) {
1165 ret
= btrfs_write_dirty_block_groups(trans
, root
);
1168 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1173 if (!list_empty(&fs_info
->dirty_cowonly_roots
))
1176 list_add_tail(&fs_info
->extent_root
->dirty_list
,
1177 &trans
->transaction
->switch_commits
);
1178 btrfs_after_dev_replace_commit(fs_info
);
1184 * dead roots are old snapshots that need to be deleted. This allocates
1185 * a dirty root struct and adds it into the list of dead roots that need to
1188 void btrfs_add_dead_root(struct btrfs_root
*root
)
1190 spin_lock(&root
->fs_info
->trans_lock
);
1191 if (list_empty(&root
->root_list
))
1192 list_add_tail(&root
->root_list
, &root
->fs_info
->dead_roots
);
1193 spin_unlock(&root
->fs_info
->trans_lock
);
1197 * update all the cowonly tree roots on disk
1199 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
,
1200 struct btrfs_root
*root
)
1202 struct btrfs_root
*gang
[8];
1203 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1208 spin_lock(&fs_info
->fs_roots_radix_lock
);
1210 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
1213 BTRFS_ROOT_TRANS_TAG
);
1216 for (i
= 0; i
< ret
; i
++) {
1218 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
1219 (unsigned long)root
->root_key
.objectid
,
1220 BTRFS_ROOT_TRANS_TAG
);
1221 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1223 btrfs_free_log(trans
, root
);
1224 btrfs_update_reloc_root(trans
, root
);
1225 btrfs_orphan_commit_root(trans
, root
);
1227 btrfs_save_ino_cache(root
, trans
);
1229 /* see comments in should_cow_block() */
1230 clear_bit(BTRFS_ROOT_FORCE_COW
, &root
->state
);
1231 smp_mb__after_atomic();
1233 if (root
->commit_root
!= root
->node
) {
1234 list_add_tail(&root
->dirty_list
,
1235 &trans
->transaction
->switch_commits
);
1236 btrfs_set_root_node(&root
->root_item
,
1240 err
= btrfs_update_root(trans
, fs_info
->tree_root
,
1243 spin_lock(&fs_info
->fs_roots_radix_lock
);
1248 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1253 * defrag a given btree.
1254 * Every leaf in the btree is read and defragged.
1256 int btrfs_defrag_root(struct btrfs_root
*root
)
1258 struct btrfs_fs_info
*info
= root
->fs_info
;
1259 struct btrfs_trans_handle
*trans
;
1262 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING
, &root
->state
))
1266 trans
= btrfs_start_transaction(root
, 0);
1268 return PTR_ERR(trans
);
1270 ret
= btrfs_defrag_leaves(trans
, root
);
1272 btrfs_end_transaction(trans
, root
);
1273 btrfs_btree_balance_dirty(info
->tree_root
);
1276 if (btrfs_fs_closing(root
->fs_info
) || ret
!= -EAGAIN
)
1279 if (btrfs_defrag_cancelled(root
->fs_info
)) {
1280 pr_debug("BTRFS: defrag_root cancelled\n");
1285 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING
, &root
->state
);
1290 * new snapshots need to be created at a very specific time in the
1291 * transaction commit. This does the actual creation.
1294 * If the error which may affect the commitment of the current transaction
1295 * happens, we should return the error number. If the error which just affect
1296 * the creation of the pending snapshots, just return 0.
1298 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
1299 struct btrfs_fs_info
*fs_info
,
1300 struct btrfs_pending_snapshot
*pending
)
1302 struct btrfs_key key
;
1303 struct btrfs_root_item
*new_root_item
;
1304 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1305 struct btrfs_root
*root
= pending
->root
;
1306 struct btrfs_root
*parent_root
;
1307 struct btrfs_block_rsv
*rsv
;
1308 struct inode
*parent_inode
;
1309 struct btrfs_path
*path
;
1310 struct btrfs_dir_item
*dir_item
;
1311 struct dentry
*dentry
;
1312 struct extent_buffer
*tmp
;
1313 struct extent_buffer
*old
;
1314 struct timespec cur_time
= CURRENT_TIME
;
1322 path
= btrfs_alloc_path();
1324 pending
->error
= -ENOMEM
;
1328 new_root_item
= kmalloc(sizeof(*new_root_item
), GFP_NOFS
);
1329 if (!new_root_item
) {
1330 pending
->error
= -ENOMEM
;
1331 goto root_item_alloc_fail
;
1334 pending
->error
= btrfs_find_free_objectid(tree_root
, &objectid
);
1336 goto no_free_objectid
;
1339 * Make qgroup to skip current new snapshot's qgroupid, as it is
1340 * accounted by later btrfs_qgroup_inherit().
1342 btrfs_set_skip_qgroup(trans
, objectid
);
1344 btrfs_reloc_pre_snapshot(pending
, &to_reserve
);
1346 if (to_reserve
> 0) {
1347 pending
->error
= btrfs_block_rsv_add(root
,
1348 &pending
->block_rsv
,
1350 BTRFS_RESERVE_NO_FLUSH
);
1352 goto clear_skip_qgroup
;
1355 key
.objectid
= objectid
;
1356 key
.offset
= (u64
)-1;
1357 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1359 rsv
= trans
->block_rsv
;
1360 trans
->block_rsv
= &pending
->block_rsv
;
1361 trans
->bytes_reserved
= trans
->block_rsv
->reserved
;
1363 dentry
= pending
->dentry
;
1364 parent_inode
= pending
->dir
;
1365 parent_root
= BTRFS_I(parent_inode
)->root
;
1366 record_root_in_trans(trans
, parent_root
);
1369 * insert the directory item
1371 ret
= btrfs_set_inode_index(parent_inode
, &index
);
1372 BUG_ON(ret
); /* -ENOMEM */
1374 /* check if there is a file/dir which has the same name. */
1375 dir_item
= btrfs_lookup_dir_item(NULL
, parent_root
, path
,
1376 btrfs_ino(parent_inode
),
1377 dentry
->d_name
.name
,
1378 dentry
->d_name
.len
, 0);
1379 if (dir_item
!= NULL
&& !IS_ERR(dir_item
)) {
1380 pending
->error
= -EEXIST
;
1381 goto dir_item_existed
;
1382 } else if (IS_ERR(dir_item
)) {
1383 ret
= PTR_ERR(dir_item
);
1384 btrfs_abort_transaction(trans
, root
, ret
);
1387 btrfs_release_path(path
);
1390 * pull in the delayed directory update
1391 * and the delayed inode item
1392 * otherwise we corrupt the FS during
1395 ret
= btrfs_run_delayed_items(trans
, root
);
1396 if (ret
) { /* Transaction aborted */
1397 btrfs_abort_transaction(trans
, root
, ret
);
1401 record_root_in_trans(trans
, root
);
1402 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
1403 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
1404 btrfs_check_and_init_root_item(new_root_item
);
1406 root_flags
= btrfs_root_flags(new_root_item
);
1407 if (pending
->readonly
)
1408 root_flags
|= BTRFS_ROOT_SUBVOL_RDONLY
;
1410 root_flags
&= ~BTRFS_ROOT_SUBVOL_RDONLY
;
1411 btrfs_set_root_flags(new_root_item
, root_flags
);
1413 btrfs_set_root_generation_v2(new_root_item
,
1415 uuid_le_gen(&new_uuid
);
1416 memcpy(new_root_item
->uuid
, new_uuid
.b
, BTRFS_UUID_SIZE
);
1417 memcpy(new_root_item
->parent_uuid
, root
->root_item
.uuid
,
1419 if (!(root_flags
& BTRFS_ROOT_SUBVOL_RDONLY
)) {
1420 memset(new_root_item
->received_uuid
, 0,
1421 sizeof(new_root_item
->received_uuid
));
1422 memset(&new_root_item
->stime
, 0, sizeof(new_root_item
->stime
));
1423 memset(&new_root_item
->rtime
, 0, sizeof(new_root_item
->rtime
));
1424 btrfs_set_root_stransid(new_root_item
, 0);
1425 btrfs_set_root_rtransid(new_root_item
, 0);
1427 btrfs_set_stack_timespec_sec(&new_root_item
->otime
, cur_time
.tv_sec
);
1428 btrfs_set_stack_timespec_nsec(&new_root_item
->otime
, cur_time
.tv_nsec
);
1429 btrfs_set_root_otransid(new_root_item
, trans
->transid
);
1431 old
= btrfs_lock_root_node(root
);
1432 ret
= btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
);
1434 btrfs_tree_unlock(old
);
1435 free_extent_buffer(old
);
1436 btrfs_abort_transaction(trans
, root
, ret
);
1440 btrfs_set_lock_blocking(old
);
1442 ret
= btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
1443 /* clean up in any case */
1444 btrfs_tree_unlock(old
);
1445 free_extent_buffer(old
);
1447 btrfs_abort_transaction(trans
, root
, ret
);
1450 /* see comments in should_cow_block() */
1451 set_bit(BTRFS_ROOT_FORCE_COW
, &root
->state
);
1454 btrfs_set_root_node(new_root_item
, tmp
);
1455 /* record when the snapshot was created in key.offset */
1456 key
.offset
= trans
->transid
;
1457 ret
= btrfs_insert_root(trans
, tree_root
, &key
, new_root_item
);
1458 btrfs_tree_unlock(tmp
);
1459 free_extent_buffer(tmp
);
1461 btrfs_abort_transaction(trans
, root
, ret
);
1466 * insert root back/forward references
1468 ret
= btrfs_add_root_ref(trans
, tree_root
, objectid
,
1469 parent_root
->root_key
.objectid
,
1470 btrfs_ino(parent_inode
), index
,
1471 dentry
->d_name
.name
, dentry
->d_name
.len
);
1473 btrfs_abort_transaction(trans
, root
, ret
);
1477 key
.offset
= (u64
)-1;
1478 pending
->snap
= btrfs_read_fs_root_no_name(root
->fs_info
, &key
);
1479 if (IS_ERR(pending
->snap
)) {
1480 ret
= PTR_ERR(pending
->snap
);
1481 btrfs_abort_transaction(trans
, root
, ret
);
1485 ret
= btrfs_reloc_post_snapshot(trans
, pending
);
1487 btrfs_abort_transaction(trans
, root
, ret
);
1491 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1493 btrfs_abort_transaction(trans
, root
, ret
);
1497 ret
= btrfs_insert_dir_item(trans
, parent_root
,
1498 dentry
->d_name
.name
, dentry
->d_name
.len
,
1500 BTRFS_FT_DIR
, index
);
1501 /* We have check then name at the beginning, so it is impossible. */
1502 BUG_ON(ret
== -EEXIST
|| ret
== -EOVERFLOW
);
1504 btrfs_abort_transaction(trans
, root
, ret
);
1508 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
1509 dentry
->d_name
.len
* 2);
1510 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
1511 ret
= btrfs_update_inode_fallback(trans
, parent_root
, parent_inode
);
1513 btrfs_abort_transaction(trans
, root
, ret
);
1516 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
, new_uuid
.b
,
1517 BTRFS_UUID_KEY_SUBVOL
, objectid
);
1519 btrfs_abort_transaction(trans
, root
, ret
);
1522 if (!btrfs_is_empty_uuid(new_root_item
->received_uuid
)) {
1523 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
1524 new_root_item
->received_uuid
,
1525 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
1527 if (ret
&& ret
!= -EEXIST
) {
1528 btrfs_abort_transaction(trans
, root
, ret
);
1533 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1535 btrfs_abort_transaction(trans
, root
, ret
);
1540 * account qgroup counters before qgroup_inherit()
1542 ret
= btrfs_qgroup_prepare_account_extents(trans
, fs_info
);
1545 ret
= btrfs_qgroup_account_extents(trans
, fs_info
);
1548 ret
= btrfs_qgroup_inherit(trans
, fs_info
,
1549 root
->root_key
.objectid
,
1550 objectid
, pending
->inherit
);
1552 btrfs_abort_transaction(trans
, root
, ret
);
1557 pending
->error
= ret
;
1559 trans
->block_rsv
= rsv
;
1560 trans
->bytes_reserved
= 0;
1562 btrfs_clear_skip_qgroup(trans
);
1564 kfree(new_root_item
);
1565 root_item_alloc_fail
:
1566 btrfs_free_path(path
);
1571 * create all the snapshots we've scheduled for creation
1573 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
,
1574 struct btrfs_fs_info
*fs_info
)
1576 struct btrfs_pending_snapshot
*pending
, *next
;
1577 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
1580 list_for_each_entry_safe(pending
, next
, head
, list
) {
1581 list_del(&pending
->list
);
1582 ret
= create_pending_snapshot(trans
, fs_info
, pending
);
1589 static void update_super_roots(struct btrfs_root
*root
)
1591 struct btrfs_root_item
*root_item
;
1592 struct btrfs_super_block
*super
;
1594 super
= root
->fs_info
->super_copy
;
1596 root_item
= &root
->fs_info
->chunk_root
->root_item
;
1597 super
->chunk_root
= root_item
->bytenr
;
1598 super
->chunk_root_generation
= root_item
->generation
;
1599 super
->chunk_root_level
= root_item
->level
;
1601 root_item
= &root
->fs_info
->tree_root
->root_item
;
1602 super
->root
= root_item
->bytenr
;
1603 super
->generation
= root_item
->generation
;
1604 super
->root_level
= root_item
->level
;
1605 if (btrfs_test_opt(root
, SPACE_CACHE
))
1606 super
->cache_generation
= root_item
->generation
;
1607 if (root
->fs_info
->update_uuid_tree_gen
)
1608 super
->uuid_tree_generation
= root_item
->generation
;
1611 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
1613 struct btrfs_transaction
*trans
;
1616 spin_lock(&info
->trans_lock
);
1617 trans
= info
->running_transaction
;
1619 ret
= (trans
->state
>= TRANS_STATE_COMMIT_START
);
1620 spin_unlock(&info
->trans_lock
);
1624 int btrfs_transaction_blocked(struct btrfs_fs_info
*info
)
1626 struct btrfs_transaction
*trans
;
1629 spin_lock(&info
->trans_lock
);
1630 trans
= info
->running_transaction
;
1632 ret
= is_transaction_blocked(trans
);
1633 spin_unlock(&info
->trans_lock
);
1638 * wait for the current transaction commit to start and block subsequent
1641 static void wait_current_trans_commit_start(struct btrfs_root
*root
,
1642 struct btrfs_transaction
*trans
)
1644 wait_event(root
->fs_info
->transaction_blocked_wait
,
1645 trans
->state
>= TRANS_STATE_COMMIT_START
||
1650 * wait for the current transaction to start and then become unblocked.
1653 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root
*root
,
1654 struct btrfs_transaction
*trans
)
1656 wait_event(root
->fs_info
->transaction_wait
,
1657 trans
->state
>= TRANS_STATE_UNBLOCKED
||
1662 * commit transactions asynchronously. once btrfs_commit_transaction_async
1663 * returns, any subsequent transaction will not be allowed to join.
1665 struct btrfs_async_commit
{
1666 struct btrfs_trans_handle
*newtrans
;
1667 struct btrfs_root
*root
;
1668 struct work_struct work
;
1671 static void do_async_commit(struct work_struct
*work
)
1673 struct btrfs_async_commit
*ac
=
1674 container_of(work
, struct btrfs_async_commit
, work
);
1677 * We've got freeze protection passed with the transaction.
1678 * Tell lockdep about it.
1680 if (ac
->newtrans
->type
& __TRANS_FREEZABLE
)
1681 __sb_writers_acquired(ac
->root
->fs_info
->sb
, SB_FREEZE_FS
);
1683 current
->journal_info
= ac
->newtrans
;
1685 btrfs_commit_transaction(ac
->newtrans
, ac
->root
);
1689 int btrfs_commit_transaction_async(struct btrfs_trans_handle
*trans
,
1690 struct btrfs_root
*root
,
1691 int wait_for_unblock
)
1693 struct btrfs_async_commit
*ac
;
1694 struct btrfs_transaction
*cur_trans
;
1696 ac
= kmalloc(sizeof(*ac
), GFP_NOFS
);
1700 INIT_WORK(&ac
->work
, do_async_commit
);
1702 ac
->newtrans
= btrfs_join_transaction(root
);
1703 if (IS_ERR(ac
->newtrans
)) {
1704 int err
= PTR_ERR(ac
->newtrans
);
1709 /* take transaction reference */
1710 cur_trans
= trans
->transaction
;
1711 atomic_inc(&cur_trans
->use_count
);
1713 btrfs_end_transaction(trans
, root
);
1716 * Tell lockdep we've released the freeze rwsem, since the
1717 * async commit thread will be the one to unlock it.
1719 if (ac
->newtrans
->type
& __TRANS_FREEZABLE
)
1720 __sb_writers_release(root
->fs_info
->sb
, SB_FREEZE_FS
);
1722 schedule_work(&ac
->work
);
1724 /* wait for transaction to start and unblock */
1725 if (wait_for_unblock
)
1726 wait_current_trans_commit_start_and_unblock(root
, cur_trans
);
1728 wait_current_trans_commit_start(root
, cur_trans
);
1730 if (current
->journal_info
== trans
)
1731 current
->journal_info
= NULL
;
1733 btrfs_put_transaction(cur_trans
);
1738 static void cleanup_transaction(struct btrfs_trans_handle
*trans
,
1739 struct btrfs_root
*root
, int err
)
1741 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1744 WARN_ON(trans
->use_count
> 1);
1746 btrfs_abort_transaction(trans
, root
, err
);
1748 spin_lock(&root
->fs_info
->trans_lock
);
1751 * If the transaction is removed from the list, it means this
1752 * transaction has been committed successfully, so it is impossible
1753 * to call the cleanup function.
1755 BUG_ON(list_empty(&cur_trans
->list
));
1757 list_del_init(&cur_trans
->list
);
1758 if (cur_trans
== root
->fs_info
->running_transaction
) {
1759 cur_trans
->state
= TRANS_STATE_COMMIT_DOING
;
1760 spin_unlock(&root
->fs_info
->trans_lock
);
1761 wait_event(cur_trans
->writer_wait
,
1762 atomic_read(&cur_trans
->num_writers
) == 1);
1764 spin_lock(&root
->fs_info
->trans_lock
);
1766 spin_unlock(&root
->fs_info
->trans_lock
);
1768 btrfs_cleanup_one_transaction(trans
->transaction
, root
);
1770 spin_lock(&root
->fs_info
->trans_lock
);
1771 if (cur_trans
== root
->fs_info
->running_transaction
)
1772 root
->fs_info
->running_transaction
= NULL
;
1773 spin_unlock(&root
->fs_info
->trans_lock
);
1775 if (trans
->type
& __TRANS_FREEZABLE
)
1776 sb_end_intwrite(root
->fs_info
->sb
);
1777 btrfs_put_transaction(cur_trans
);
1778 btrfs_put_transaction(cur_trans
);
1780 trace_btrfs_transaction_commit(root
);
1782 if (current
->journal_info
== trans
)
1783 current
->journal_info
= NULL
;
1784 btrfs_scrub_cancel(root
->fs_info
);
1786 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1789 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info
*fs_info
)
1791 if (btrfs_test_opt(fs_info
->tree_root
, FLUSHONCOMMIT
))
1792 return btrfs_start_delalloc_roots(fs_info
, 1, -1);
1796 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info
*fs_info
)
1798 if (btrfs_test_opt(fs_info
->tree_root
, FLUSHONCOMMIT
))
1799 btrfs_wait_ordered_roots(fs_info
, -1);
1803 btrfs_wait_pending_ordered(struct btrfs_transaction
*cur_trans
,
1804 struct btrfs_fs_info
*fs_info
)
1806 struct btrfs_ordered_extent
*ordered
;
1808 spin_lock(&fs_info
->trans_lock
);
1809 while (!list_empty(&cur_trans
->pending_ordered
)) {
1810 ordered
= list_first_entry(&cur_trans
->pending_ordered
,
1811 struct btrfs_ordered_extent
,
1813 list_del_init(&ordered
->trans_list
);
1814 spin_unlock(&fs_info
->trans_lock
);
1816 wait_event(ordered
->wait
, test_bit(BTRFS_ORDERED_COMPLETE
,
1818 btrfs_put_ordered_extent(ordered
);
1819 spin_lock(&fs_info
->trans_lock
);
1821 spin_unlock(&fs_info
->trans_lock
);
1824 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
,
1825 struct btrfs_root
*root
)
1827 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1828 struct btrfs_transaction
*prev_trans
= NULL
;
1829 struct btrfs_inode
*btree_ino
= BTRFS_I(root
->fs_info
->btree_inode
);
1832 /* Stop the commit early if ->aborted is set */
1833 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1834 ret
= cur_trans
->aborted
;
1835 btrfs_end_transaction(trans
, root
);
1839 /* make a pass through all the delayed refs we have so far
1840 * any runnings procs may add more while we are here
1842 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1844 btrfs_end_transaction(trans
, root
);
1848 btrfs_trans_release_metadata(trans
, root
);
1849 trans
->block_rsv
= NULL
;
1850 if (trans
->qgroup_reserved
) {
1851 btrfs_qgroup_free(root
, trans
->qgroup_reserved
);
1852 trans
->qgroup_reserved
= 0;
1855 cur_trans
= trans
->transaction
;
1858 * set the flushing flag so procs in this transaction have to
1859 * start sending their work down.
1861 cur_trans
->delayed_refs
.flushing
= 1;
1864 if (!list_empty(&trans
->new_bgs
))
1865 btrfs_create_pending_block_groups(trans
, root
);
1867 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1869 btrfs_end_transaction(trans
, root
);
1873 if (!cur_trans
->dirty_bg_run
) {
1876 /* this mutex is also taken before trying to set
1877 * block groups readonly. We need to make sure
1878 * that nobody has set a block group readonly
1879 * after a extents from that block group have been
1880 * allocated for cache files. btrfs_set_block_group_ro
1881 * will wait for the transaction to commit if it
1882 * finds dirty_bg_run = 1
1884 * The dirty_bg_run flag is also used to make sure only
1885 * one process starts all the block group IO. It wouldn't
1886 * hurt to have more than one go through, but there's no
1887 * real advantage to it either.
1889 mutex_lock(&root
->fs_info
->ro_block_group_mutex
);
1890 if (!cur_trans
->dirty_bg_run
) {
1892 cur_trans
->dirty_bg_run
= 1;
1894 mutex_unlock(&root
->fs_info
->ro_block_group_mutex
);
1897 ret
= btrfs_start_dirty_block_groups(trans
, root
);
1900 btrfs_end_transaction(trans
, root
);
1904 spin_lock(&root
->fs_info
->trans_lock
);
1905 list_splice_init(&trans
->ordered
, &cur_trans
->pending_ordered
);
1906 if (cur_trans
->state
>= TRANS_STATE_COMMIT_START
) {
1907 spin_unlock(&root
->fs_info
->trans_lock
);
1908 atomic_inc(&cur_trans
->use_count
);
1909 ret
= btrfs_end_transaction(trans
, root
);
1911 wait_for_commit(root
, cur_trans
);
1913 if (unlikely(cur_trans
->aborted
))
1914 ret
= cur_trans
->aborted
;
1916 btrfs_put_transaction(cur_trans
);
1921 cur_trans
->state
= TRANS_STATE_COMMIT_START
;
1922 wake_up(&root
->fs_info
->transaction_blocked_wait
);
1924 if (cur_trans
->list
.prev
!= &root
->fs_info
->trans_list
) {
1925 prev_trans
= list_entry(cur_trans
->list
.prev
,
1926 struct btrfs_transaction
, list
);
1927 if (prev_trans
->state
!= TRANS_STATE_COMPLETED
) {
1928 atomic_inc(&prev_trans
->use_count
);
1929 spin_unlock(&root
->fs_info
->trans_lock
);
1931 wait_for_commit(root
, prev_trans
);
1932 ret
= prev_trans
->aborted
;
1934 btrfs_put_transaction(prev_trans
);
1936 goto cleanup_transaction
;
1938 spin_unlock(&root
->fs_info
->trans_lock
);
1941 spin_unlock(&root
->fs_info
->trans_lock
);
1944 extwriter_counter_dec(cur_trans
, trans
->type
);
1946 ret
= btrfs_start_delalloc_flush(root
->fs_info
);
1948 goto cleanup_transaction
;
1950 ret
= btrfs_run_delayed_items(trans
, root
);
1952 goto cleanup_transaction
;
1954 wait_event(cur_trans
->writer_wait
,
1955 extwriter_counter_read(cur_trans
) == 0);
1957 /* some pending stuffs might be added after the previous flush. */
1958 ret
= btrfs_run_delayed_items(trans
, root
);
1960 goto cleanup_transaction
;
1962 btrfs_wait_delalloc_flush(root
->fs_info
);
1964 btrfs_wait_pending_ordered(cur_trans
, root
->fs_info
);
1966 btrfs_scrub_pause(root
);
1968 * Ok now we need to make sure to block out any other joins while we
1969 * commit the transaction. We could have started a join before setting
1970 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
1972 spin_lock(&root
->fs_info
->trans_lock
);
1973 cur_trans
->state
= TRANS_STATE_COMMIT_DOING
;
1974 spin_unlock(&root
->fs_info
->trans_lock
);
1975 wait_event(cur_trans
->writer_wait
,
1976 atomic_read(&cur_trans
->num_writers
) == 1);
1978 /* ->aborted might be set after the previous check, so check it */
1979 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1980 ret
= cur_trans
->aborted
;
1981 goto scrub_continue
;
1984 * the reloc mutex makes sure that we stop
1985 * the balancing code from coming in and moving
1986 * extents around in the middle of the commit
1988 mutex_lock(&root
->fs_info
->reloc_mutex
);
1991 * We needn't worry about the delayed items because we will
1992 * deal with them in create_pending_snapshot(), which is the
1993 * core function of the snapshot creation.
1995 ret
= create_pending_snapshots(trans
, root
->fs_info
);
1997 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1998 goto scrub_continue
;
2002 * We insert the dir indexes of the snapshots and update the inode
2003 * of the snapshots' parents after the snapshot creation, so there
2004 * are some delayed items which are not dealt with. Now deal with
2007 * We needn't worry that this operation will corrupt the snapshots,
2008 * because all the tree which are snapshoted will be forced to COW
2009 * the nodes and leaves.
2011 ret
= btrfs_run_delayed_items(trans
, root
);
2013 mutex_unlock(&root
->fs_info
->reloc_mutex
);
2014 goto scrub_continue
;
2017 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
2019 mutex_unlock(&root
->fs_info
->reloc_mutex
);
2020 goto scrub_continue
;
2023 /* Reocrd old roots for later qgroup accounting */
2024 ret
= btrfs_qgroup_prepare_account_extents(trans
, root
->fs_info
);
2026 mutex_unlock(&root
->fs_info
->reloc_mutex
);
2027 goto scrub_continue
;
2031 * make sure none of the code above managed to slip in a
2034 btrfs_assert_delayed_root_empty(root
);
2036 WARN_ON(cur_trans
!= trans
->transaction
);
2038 /* btrfs_commit_tree_roots is responsible for getting the
2039 * various roots consistent with each other. Every pointer
2040 * in the tree of tree roots has to point to the most up to date
2041 * root for every subvolume and other tree. So, we have to keep
2042 * the tree logging code from jumping in and changing any
2045 * At this point in the commit, there can't be any tree-log
2046 * writers, but a little lower down we drop the trans mutex
2047 * and let new people in. By holding the tree_log_mutex
2048 * from now until after the super is written, we avoid races
2049 * with the tree-log code.
2051 mutex_lock(&root
->fs_info
->tree_log_mutex
);
2053 ret
= commit_fs_roots(trans
, root
);
2055 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
2056 mutex_unlock(&root
->fs_info
->reloc_mutex
);
2057 goto scrub_continue
;
2061 * Since the transaction is done, we can apply the pending changes
2062 * before the next transaction.
2064 btrfs_apply_pending_changes(root
->fs_info
);
2066 /* commit_fs_roots gets rid of all the tree log roots, it is now
2067 * safe to free the root of tree log roots
2069 btrfs_free_log_root_tree(trans
, root
->fs_info
);
2072 * Since fs roots are all committed, we can get a quite accurate
2073 * new_roots. So let's do quota accounting.
2075 ret
= btrfs_qgroup_account_extents(trans
, root
->fs_info
);
2077 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
2078 mutex_unlock(&root
->fs_info
->reloc_mutex
);
2079 goto scrub_continue
;
2082 ret
= commit_cowonly_roots(trans
, root
);
2084 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
2085 mutex_unlock(&root
->fs_info
->reloc_mutex
);
2086 goto scrub_continue
;
2090 * The tasks which save the space cache and inode cache may also
2091 * update ->aborted, check it.
2093 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
2094 ret
= cur_trans
->aborted
;
2095 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
2096 mutex_unlock(&root
->fs_info
->reloc_mutex
);
2097 goto scrub_continue
;
2100 btrfs_prepare_extent_commit(trans
, root
);
2102 cur_trans
= root
->fs_info
->running_transaction
;
2104 btrfs_set_root_node(&root
->fs_info
->tree_root
->root_item
,
2105 root
->fs_info
->tree_root
->node
);
2106 list_add_tail(&root
->fs_info
->tree_root
->dirty_list
,
2107 &cur_trans
->switch_commits
);
2109 btrfs_set_root_node(&root
->fs_info
->chunk_root
->root_item
,
2110 root
->fs_info
->chunk_root
->node
);
2111 list_add_tail(&root
->fs_info
->chunk_root
->dirty_list
,
2112 &cur_trans
->switch_commits
);
2114 switch_commit_roots(cur_trans
, root
->fs_info
);
2116 assert_qgroups_uptodate(trans
);
2117 ASSERT(list_empty(&cur_trans
->dirty_bgs
));
2118 ASSERT(list_empty(&cur_trans
->io_bgs
));
2119 update_super_roots(root
);
2121 btrfs_set_super_log_root(root
->fs_info
->super_copy
, 0);
2122 btrfs_set_super_log_root_level(root
->fs_info
->super_copy
, 0);
2123 memcpy(root
->fs_info
->super_for_commit
, root
->fs_info
->super_copy
,
2124 sizeof(*root
->fs_info
->super_copy
));
2126 btrfs_update_commit_device_size(root
->fs_info
);
2127 btrfs_update_commit_device_bytes_used(root
, cur_trans
);
2129 clear_bit(BTRFS_INODE_BTREE_LOG1_ERR
, &btree_ino
->runtime_flags
);
2130 clear_bit(BTRFS_INODE_BTREE_LOG2_ERR
, &btree_ino
->runtime_flags
);
2132 btrfs_trans_release_chunk_metadata(trans
);
2134 spin_lock(&root
->fs_info
->trans_lock
);
2135 cur_trans
->state
= TRANS_STATE_UNBLOCKED
;
2136 root
->fs_info
->running_transaction
= NULL
;
2137 spin_unlock(&root
->fs_info
->trans_lock
);
2138 mutex_unlock(&root
->fs_info
->reloc_mutex
);
2140 wake_up(&root
->fs_info
->transaction_wait
);
2142 ret
= btrfs_write_and_wait_transaction(trans
, root
);
2144 btrfs_error(root
->fs_info
, ret
,
2145 "Error while writing out transaction");
2146 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
2147 goto scrub_continue
;
2150 ret
= write_ctree_super(trans
, root
, 0);
2152 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
2153 goto scrub_continue
;
2157 * the super is written, we can safely allow the tree-loggers
2158 * to go about their business
2160 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
2162 btrfs_finish_extent_commit(trans
, root
);
2164 if (cur_trans
->have_free_bgs
)
2165 btrfs_clear_space_info_full(root
->fs_info
);
2167 root
->fs_info
->last_trans_committed
= cur_trans
->transid
;
2169 * We needn't acquire the lock here because there is no other task
2170 * which can change it.
2172 cur_trans
->state
= TRANS_STATE_COMPLETED
;
2173 wake_up(&cur_trans
->commit_wait
);
2175 spin_lock(&root
->fs_info
->trans_lock
);
2176 list_del_init(&cur_trans
->list
);
2177 spin_unlock(&root
->fs_info
->trans_lock
);
2179 btrfs_put_transaction(cur_trans
);
2180 btrfs_put_transaction(cur_trans
);
2182 if (trans
->type
& __TRANS_FREEZABLE
)
2183 sb_end_intwrite(root
->fs_info
->sb
);
2185 trace_btrfs_transaction_commit(root
);
2187 btrfs_scrub_continue(root
);
2189 if (current
->journal_info
== trans
)
2190 current
->journal_info
= NULL
;
2192 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
2194 if (current
!= root
->fs_info
->transaction_kthread
&&
2195 current
!= root
->fs_info
->cleaner_kthread
)
2196 btrfs_run_delayed_iputs(root
);
2201 btrfs_scrub_continue(root
);
2202 cleanup_transaction
:
2203 btrfs_trans_release_metadata(trans
, root
);
2204 btrfs_trans_release_chunk_metadata(trans
);
2205 trans
->block_rsv
= NULL
;
2206 if (trans
->qgroup_reserved
) {
2207 btrfs_qgroup_free(root
, trans
->qgroup_reserved
);
2208 trans
->qgroup_reserved
= 0;
2210 btrfs_warn(root
->fs_info
, "Skipping commit of aborted transaction.");
2211 if (current
->journal_info
== trans
)
2212 current
->journal_info
= NULL
;
2213 cleanup_transaction(trans
, root
, ret
);
2219 * return < 0 if error
2220 * 0 if there are no more dead_roots at the time of call
2221 * 1 there are more to be processed, call me again
2223 * The return value indicates there are certainly more snapshots to delete, but
2224 * if there comes a new one during processing, it may return 0. We don't mind,
2225 * because btrfs_commit_super will poke cleaner thread and it will process it a
2226 * few seconds later.
2228 int btrfs_clean_one_deleted_snapshot(struct btrfs_root
*root
)
2231 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2233 spin_lock(&fs_info
->trans_lock
);
2234 if (list_empty(&fs_info
->dead_roots
)) {
2235 spin_unlock(&fs_info
->trans_lock
);
2238 root
= list_first_entry(&fs_info
->dead_roots
,
2239 struct btrfs_root
, root_list
);
2240 list_del_init(&root
->root_list
);
2241 spin_unlock(&fs_info
->trans_lock
);
2243 pr_debug("BTRFS: cleaner removing %llu\n", root
->objectid
);
2245 btrfs_kill_all_delayed_nodes(root
);
2247 if (btrfs_header_backref_rev(root
->node
) <
2248 BTRFS_MIXED_BACKREF_REV
)
2249 ret
= btrfs_drop_snapshot(root
, NULL
, 0, 0);
2251 ret
= btrfs_drop_snapshot(root
, NULL
, 1, 0);
2253 return (ret
< 0) ? 0 : 1;
2256 void btrfs_apply_pending_changes(struct btrfs_fs_info
*fs_info
)
2261 prev
= xchg(&fs_info
->pending_changes
, 0);
2265 bit
= 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE
;
2267 btrfs_set_opt(fs_info
->mount_opt
, INODE_MAP_CACHE
);
2270 bit
= 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE
;
2272 btrfs_clear_opt(fs_info
->mount_opt
, INODE_MAP_CACHE
);
2275 bit
= 1 << BTRFS_PENDING_COMMIT
;
2277 btrfs_debug(fs_info
, "pending commit done");
2282 "unknown pending changes left 0x%lx, ignoring", prev
);