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(refcount_read(&transaction
->use_count
) == 0);
64 if (refcount_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 btrfs_err(transaction
->fs_info
,
69 "pending csums is %llu",
70 transaction
->delayed_refs
.pending_csums
);
71 while (!list_empty(&transaction
->pending_chunks
)) {
72 struct extent_map
*em
;
74 em
= list_first_entry(&transaction
->pending_chunks
,
75 struct extent_map
, list
);
76 list_del_init(&em
->list
);
80 * If any block groups are found in ->deleted_bgs then it's
81 * because the transaction was aborted and a commit did not
82 * happen (things failed before writing the new superblock
83 * and calling btrfs_finish_extent_commit()), so we can not
84 * discard the physical locations of the block groups.
86 while (!list_empty(&transaction
->deleted_bgs
)) {
87 struct btrfs_block_group_cache
*cache
;
89 cache
= list_first_entry(&transaction
->deleted_bgs
,
90 struct btrfs_block_group_cache
,
92 list_del_init(&cache
->bg_list
);
93 btrfs_put_block_group_trimming(cache
);
94 btrfs_put_block_group(cache
);
100 static void clear_btree_io_tree(struct extent_io_tree
*tree
)
102 spin_lock(&tree
->lock
);
104 * Do a single barrier for the waitqueue_active check here, the state
105 * of the waitqueue should not change once clear_btree_io_tree is
109 while (!RB_EMPTY_ROOT(&tree
->state
)) {
110 struct rb_node
*node
;
111 struct extent_state
*state
;
113 node
= rb_first(&tree
->state
);
114 state
= rb_entry(node
, struct extent_state
, rb_node
);
115 rb_erase(&state
->rb_node
, &tree
->state
);
116 RB_CLEAR_NODE(&state
->rb_node
);
118 * btree io trees aren't supposed to have tasks waiting for
119 * changes in the flags of extent states ever.
121 ASSERT(!waitqueue_active(&state
->wq
));
122 free_extent_state(state
);
124 cond_resched_lock(&tree
->lock
);
126 spin_unlock(&tree
->lock
);
129 static noinline
void switch_commit_roots(struct btrfs_transaction
*trans
,
130 struct btrfs_fs_info
*fs_info
)
132 struct btrfs_root
*root
, *tmp
;
134 down_write(&fs_info
->commit_root_sem
);
135 list_for_each_entry_safe(root
, tmp
, &trans
->switch_commits
,
137 list_del_init(&root
->dirty_list
);
138 free_extent_buffer(root
->commit_root
);
139 root
->commit_root
= btrfs_root_node(root
);
140 if (is_fstree(root
->objectid
))
141 btrfs_unpin_free_ino(root
);
142 clear_btree_io_tree(&root
->dirty_log_pages
);
145 /* We can free old roots now. */
146 spin_lock(&trans
->dropped_roots_lock
);
147 while (!list_empty(&trans
->dropped_roots
)) {
148 root
= list_first_entry(&trans
->dropped_roots
,
149 struct btrfs_root
, root_list
);
150 list_del_init(&root
->root_list
);
151 spin_unlock(&trans
->dropped_roots_lock
);
152 btrfs_drop_and_free_fs_root(fs_info
, root
);
153 spin_lock(&trans
->dropped_roots_lock
);
155 spin_unlock(&trans
->dropped_roots_lock
);
156 up_write(&fs_info
->commit_root_sem
);
159 static inline void extwriter_counter_inc(struct btrfs_transaction
*trans
,
162 if (type
& TRANS_EXTWRITERS
)
163 atomic_inc(&trans
->num_extwriters
);
166 static inline void extwriter_counter_dec(struct btrfs_transaction
*trans
,
169 if (type
& TRANS_EXTWRITERS
)
170 atomic_dec(&trans
->num_extwriters
);
173 static inline void extwriter_counter_init(struct btrfs_transaction
*trans
,
176 atomic_set(&trans
->num_extwriters
, ((type
& TRANS_EXTWRITERS
) ? 1 : 0));
179 static inline int extwriter_counter_read(struct btrfs_transaction
*trans
)
181 return atomic_read(&trans
->num_extwriters
);
185 * either allocate a new transaction or hop into the existing one
187 static noinline
int join_transaction(struct btrfs_fs_info
*fs_info
,
190 struct btrfs_transaction
*cur_trans
;
192 spin_lock(&fs_info
->trans_lock
);
194 /* The file system has been taken offline. No new transactions. */
195 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
196 spin_unlock(&fs_info
->trans_lock
);
200 cur_trans
= fs_info
->running_transaction
;
202 if (cur_trans
->aborted
) {
203 spin_unlock(&fs_info
->trans_lock
);
204 return cur_trans
->aborted
;
206 if (btrfs_blocked_trans_types
[cur_trans
->state
] & type
) {
207 spin_unlock(&fs_info
->trans_lock
);
210 refcount_inc(&cur_trans
->use_count
);
211 atomic_inc(&cur_trans
->num_writers
);
212 extwriter_counter_inc(cur_trans
, type
);
213 spin_unlock(&fs_info
->trans_lock
);
216 spin_unlock(&fs_info
->trans_lock
);
219 * If we are ATTACH, we just want to catch the current transaction,
220 * and commit it. If there is no transaction, just return ENOENT.
222 if (type
== TRANS_ATTACH
)
226 * JOIN_NOLOCK only happens during the transaction commit, so
227 * it is impossible that ->running_transaction is NULL
229 BUG_ON(type
== TRANS_JOIN_NOLOCK
);
231 cur_trans
= kmalloc(sizeof(*cur_trans
), GFP_NOFS
);
235 spin_lock(&fs_info
->trans_lock
);
236 if (fs_info
->running_transaction
) {
238 * someone started a transaction after we unlocked. Make sure
239 * to redo the checks above
243 } else if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
244 spin_unlock(&fs_info
->trans_lock
);
249 cur_trans
->fs_info
= fs_info
;
250 atomic_set(&cur_trans
->num_writers
, 1);
251 extwriter_counter_init(cur_trans
, type
);
252 init_waitqueue_head(&cur_trans
->writer_wait
);
253 init_waitqueue_head(&cur_trans
->commit_wait
);
254 init_waitqueue_head(&cur_trans
->pending_wait
);
255 cur_trans
->state
= TRANS_STATE_RUNNING
;
257 * One for this trans handle, one so it will live on until we
258 * commit the transaction.
260 refcount_set(&cur_trans
->use_count
, 2);
261 atomic_set(&cur_trans
->pending_ordered
, 0);
262 cur_trans
->flags
= 0;
263 cur_trans
->start_time
= get_seconds();
265 memset(&cur_trans
->delayed_refs
, 0, sizeof(cur_trans
->delayed_refs
));
267 cur_trans
->delayed_refs
.href_root
= RB_ROOT
;
268 cur_trans
->delayed_refs
.dirty_extent_root
= RB_ROOT
;
269 atomic_set(&cur_trans
->delayed_refs
.num_entries
, 0);
272 * although the tree mod log is per file system and not per transaction,
273 * the log must never go across transaction boundaries.
276 if (!list_empty(&fs_info
->tree_mod_seq_list
))
277 WARN(1, KERN_ERR
"BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n");
278 if (!RB_EMPTY_ROOT(&fs_info
->tree_mod_log
))
279 WARN(1, KERN_ERR
"BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n");
280 atomic64_set(&fs_info
->tree_mod_seq
, 0);
282 spin_lock_init(&cur_trans
->delayed_refs
.lock
);
284 INIT_LIST_HEAD(&cur_trans
->pending_snapshots
);
285 INIT_LIST_HEAD(&cur_trans
->pending_chunks
);
286 INIT_LIST_HEAD(&cur_trans
->switch_commits
);
287 INIT_LIST_HEAD(&cur_trans
->dirty_bgs
);
288 INIT_LIST_HEAD(&cur_trans
->io_bgs
);
289 INIT_LIST_HEAD(&cur_trans
->dropped_roots
);
290 mutex_init(&cur_trans
->cache_write_mutex
);
291 cur_trans
->num_dirty_bgs
= 0;
292 spin_lock_init(&cur_trans
->dirty_bgs_lock
);
293 INIT_LIST_HEAD(&cur_trans
->deleted_bgs
);
294 spin_lock_init(&cur_trans
->dropped_roots_lock
);
295 list_add_tail(&cur_trans
->list
, &fs_info
->trans_list
);
296 extent_io_tree_init(&cur_trans
->dirty_pages
,
297 fs_info
->btree_inode
);
298 fs_info
->generation
++;
299 cur_trans
->transid
= fs_info
->generation
;
300 fs_info
->running_transaction
= cur_trans
;
301 cur_trans
->aborted
= 0;
302 spin_unlock(&fs_info
->trans_lock
);
308 * this does all the record keeping required to make sure that a reference
309 * counted root is properly recorded in a given transaction. This is required
310 * to make sure the old root from before we joined the transaction is deleted
311 * when the transaction commits
313 static int record_root_in_trans(struct btrfs_trans_handle
*trans
,
314 struct btrfs_root
*root
,
317 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
319 if ((test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) &&
320 root
->last_trans
< trans
->transid
) || force
) {
321 WARN_ON(root
== fs_info
->extent_root
);
322 WARN_ON(!force
&& root
->commit_root
!= root
->node
);
325 * see below for IN_TRANS_SETUP usage rules
326 * we have the reloc mutex held now, so there
327 * is only one writer in this function
329 set_bit(BTRFS_ROOT_IN_TRANS_SETUP
, &root
->state
);
331 /* make sure readers find IN_TRANS_SETUP before
332 * they find our root->last_trans update
336 spin_lock(&fs_info
->fs_roots_radix_lock
);
337 if (root
->last_trans
== trans
->transid
&& !force
) {
338 spin_unlock(&fs_info
->fs_roots_radix_lock
);
341 radix_tree_tag_set(&fs_info
->fs_roots_radix
,
342 (unsigned long)root
->root_key
.objectid
,
343 BTRFS_ROOT_TRANS_TAG
);
344 spin_unlock(&fs_info
->fs_roots_radix_lock
);
345 root
->last_trans
= trans
->transid
;
347 /* this is pretty tricky. We don't want to
348 * take the relocation lock in btrfs_record_root_in_trans
349 * unless we're really doing the first setup for this root in
352 * Normally we'd use root->last_trans as a flag to decide
353 * if we want to take the expensive mutex.
355 * But, we have to set root->last_trans before we
356 * init the relocation root, otherwise, we trip over warnings
357 * in ctree.c. The solution used here is to flag ourselves
358 * with root IN_TRANS_SETUP. When this is 1, we're still
359 * fixing up the reloc trees and everyone must wait.
361 * When this is zero, they can trust root->last_trans and fly
362 * through btrfs_record_root_in_trans without having to take the
363 * lock. smp_wmb() makes sure that all the writes above are
364 * done before we pop in the zero below
366 btrfs_init_reloc_root(trans
, root
);
367 smp_mb__before_atomic();
368 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP
, &root
->state
);
374 void btrfs_add_dropped_root(struct btrfs_trans_handle
*trans
,
375 struct btrfs_root
*root
)
377 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
378 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
380 /* Add ourselves to the transaction dropped list */
381 spin_lock(&cur_trans
->dropped_roots_lock
);
382 list_add_tail(&root
->root_list
, &cur_trans
->dropped_roots
);
383 spin_unlock(&cur_trans
->dropped_roots_lock
);
385 /* Make sure we don't try to update the root at commit time */
386 spin_lock(&fs_info
->fs_roots_radix_lock
);
387 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
388 (unsigned long)root
->root_key
.objectid
,
389 BTRFS_ROOT_TRANS_TAG
);
390 spin_unlock(&fs_info
->fs_roots_radix_lock
);
393 int btrfs_record_root_in_trans(struct btrfs_trans_handle
*trans
,
394 struct btrfs_root
*root
)
396 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
398 if (!test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
402 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
406 if (root
->last_trans
== trans
->transid
&&
407 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP
, &root
->state
))
410 mutex_lock(&fs_info
->reloc_mutex
);
411 record_root_in_trans(trans
, root
, 0);
412 mutex_unlock(&fs_info
->reloc_mutex
);
417 static inline int is_transaction_blocked(struct btrfs_transaction
*trans
)
419 return (trans
->state
>= TRANS_STATE_BLOCKED
&&
420 trans
->state
< TRANS_STATE_UNBLOCKED
&&
424 /* wait for commit against the current transaction to become unblocked
425 * when this is done, it is safe to start a new transaction, but the current
426 * transaction might not be fully on disk.
428 static void wait_current_trans(struct btrfs_fs_info
*fs_info
)
430 struct btrfs_transaction
*cur_trans
;
432 spin_lock(&fs_info
->trans_lock
);
433 cur_trans
= fs_info
->running_transaction
;
434 if (cur_trans
&& is_transaction_blocked(cur_trans
)) {
435 refcount_inc(&cur_trans
->use_count
);
436 spin_unlock(&fs_info
->trans_lock
);
438 wait_event(fs_info
->transaction_wait
,
439 cur_trans
->state
>= TRANS_STATE_UNBLOCKED
||
441 btrfs_put_transaction(cur_trans
);
443 spin_unlock(&fs_info
->trans_lock
);
447 static int may_wait_transaction(struct btrfs_fs_info
*fs_info
, int type
)
449 if (test_bit(BTRFS_FS_LOG_RECOVERING
, &fs_info
->flags
))
452 if (type
== TRANS_USERSPACE
)
455 if (type
== TRANS_START
&&
456 !atomic_read(&fs_info
->open_ioctl_trans
))
462 static inline bool need_reserve_reloc_root(struct btrfs_root
*root
)
464 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
466 if (!fs_info
->reloc_ctl
||
467 !test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
468 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
475 static struct btrfs_trans_handle
*
476 start_transaction(struct btrfs_root
*root
, unsigned int num_items
,
477 unsigned int type
, enum btrfs_reserve_flush_enum flush
,
478 bool enforce_qgroups
)
480 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
482 struct btrfs_trans_handle
*h
;
483 struct btrfs_transaction
*cur_trans
;
485 u64 qgroup_reserved
= 0;
486 bool reloc_reserved
= false;
489 /* Send isn't supposed to start transactions. */
490 ASSERT(current
->journal_info
!= BTRFS_SEND_TRANS_STUB
);
492 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
493 return ERR_PTR(-EROFS
);
495 if (current
->journal_info
) {
496 WARN_ON(type
& TRANS_EXTWRITERS
);
497 h
= current
->journal_info
;
499 WARN_ON(h
->use_count
> 2);
500 h
->orig_rsv
= h
->block_rsv
;
506 * Do the reservation before we join the transaction so we can do all
507 * the appropriate flushing if need be.
509 if (num_items
&& root
!= fs_info
->chunk_root
) {
510 qgroup_reserved
= num_items
* fs_info
->nodesize
;
511 ret
= btrfs_qgroup_reserve_meta(root
, qgroup_reserved
,
516 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, num_items
);
518 * Do the reservation for the relocation root creation
520 if (need_reserve_reloc_root(root
)) {
521 num_bytes
+= fs_info
->nodesize
;
522 reloc_reserved
= true;
525 ret
= btrfs_block_rsv_add(root
, &fs_info
->trans_block_rsv
,
531 h
= kmem_cache_zalloc(btrfs_trans_handle_cachep
, GFP_NOFS
);
538 * If we are JOIN_NOLOCK we're already committing a transaction and
539 * waiting on this guy, so we don't need to do the sb_start_intwrite
540 * because we're already holding a ref. We need this because we could
541 * have raced in and did an fsync() on a file which can kick a commit
542 * and then we deadlock with somebody doing a freeze.
544 * If we are ATTACH, it means we just want to catch the current
545 * transaction and commit it, so we needn't do sb_start_intwrite().
547 if (type
& __TRANS_FREEZABLE
)
548 sb_start_intwrite(fs_info
->sb
);
550 if (may_wait_transaction(fs_info
, type
))
551 wait_current_trans(fs_info
);
554 ret
= join_transaction(fs_info
, type
);
556 wait_current_trans(fs_info
);
557 if (unlikely(type
== TRANS_ATTACH
))
560 } while (ret
== -EBUSY
);
565 cur_trans
= fs_info
->running_transaction
;
567 h
->transid
= cur_trans
->transid
;
568 h
->transaction
= cur_trans
;
571 h
->fs_info
= root
->fs_info
;
574 h
->can_flush_pending_bgs
= true;
575 INIT_LIST_HEAD(&h
->new_bgs
);
578 if (cur_trans
->state
>= TRANS_STATE_BLOCKED
&&
579 may_wait_transaction(fs_info
, type
)) {
580 current
->journal_info
= h
;
581 btrfs_commit_transaction(h
);
586 trace_btrfs_space_reservation(fs_info
, "transaction",
587 h
->transid
, num_bytes
, 1);
588 h
->block_rsv
= &fs_info
->trans_block_rsv
;
589 h
->bytes_reserved
= num_bytes
;
590 h
->reloc_reserved
= reloc_reserved
;
594 btrfs_record_root_in_trans(h
, root
);
596 if (!current
->journal_info
&& type
!= TRANS_USERSPACE
)
597 current
->journal_info
= h
;
601 if (type
& __TRANS_FREEZABLE
)
602 sb_end_intwrite(fs_info
->sb
);
603 kmem_cache_free(btrfs_trans_handle_cachep
, h
);
606 btrfs_block_rsv_release(fs_info
, &fs_info
->trans_block_rsv
,
609 btrfs_qgroup_free_meta(root
, qgroup_reserved
);
613 struct btrfs_trans_handle
*btrfs_start_transaction(struct btrfs_root
*root
,
614 unsigned int num_items
)
616 return start_transaction(root
, num_items
, TRANS_START
,
617 BTRFS_RESERVE_FLUSH_ALL
, true);
620 struct btrfs_trans_handle
*btrfs_start_transaction_fallback_global_rsv(
621 struct btrfs_root
*root
,
622 unsigned int num_items
,
625 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
626 struct btrfs_trans_handle
*trans
;
631 * We have two callers: unlink and block group removal. The
632 * former should succeed even if we will temporarily exceed
633 * quota and the latter operates on the extent root so
634 * qgroup enforcement is ignored anyway.
636 trans
= start_transaction(root
, num_items
, TRANS_START
,
637 BTRFS_RESERVE_FLUSH_ALL
, false);
638 if (!IS_ERR(trans
) || PTR_ERR(trans
) != -ENOSPC
)
641 trans
= btrfs_start_transaction(root
, 0);
645 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, num_items
);
646 ret
= btrfs_cond_migrate_bytes(fs_info
, &fs_info
->trans_block_rsv
,
647 num_bytes
, min_factor
);
649 btrfs_end_transaction(trans
);
653 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
654 trans
->bytes_reserved
= num_bytes
;
655 trace_btrfs_space_reservation(fs_info
, "transaction",
656 trans
->transid
, num_bytes
, 1);
661 struct btrfs_trans_handle
*btrfs_start_transaction_lflush(
662 struct btrfs_root
*root
,
663 unsigned int num_items
)
665 return start_transaction(root
, num_items
, TRANS_START
,
666 BTRFS_RESERVE_FLUSH_LIMIT
, true);
669 struct btrfs_trans_handle
*btrfs_join_transaction(struct btrfs_root
*root
)
671 return start_transaction(root
, 0, TRANS_JOIN
, BTRFS_RESERVE_NO_FLUSH
,
675 struct btrfs_trans_handle
*btrfs_join_transaction_nolock(struct btrfs_root
*root
)
677 return start_transaction(root
, 0, TRANS_JOIN_NOLOCK
,
678 BTRFS_RESERVE_NO_FLUSH
, true);
681 struct btrfs_trans_handle
*btrfs_start_ioctl_transaction(struct btrfs_root
*root
)
683 return start_transaction(root
, 0, TRANS_USERSPACE
,
684 BTRFS_RESERVE_NO_FLUSH
, true);
688 * btrfs_attach_transaction() - catch the running transaction
690 * It is used when we want to commit the current the transaction, but
691 * don't want to start a new one.
693 * Note: If this function return -ENOENT, it just means there is no
694 * running transaction. But it is possible that the inactive transaction
695 * is still in the memory, not fully on disk. If you hope there is no
696 * inactive transaction in the fs when -ENOENT is returned, you should
698 * btrfs_attach_transaction_barrier()
700 struct btrfs_trans_handle
*btrfs_attach_transaction(struct btrfs_root
*root
)
702 return start_transaction(root
, 0, TRANS_ATTACH
,
703 BTRFS_RESERVE_NO_FLUSH
, true);
707 * btrfs_attach_transaction_barrier() - catch the running transaction
709 * It is similar to the above function, the differentia is this one
710 * will wait for all the inactive transactions until they fully
713 struct btrfs_trans_handle
*
714 btrfs_attach_transaction_barrier(struct btrfs_root
*root
)
716 struct btrfs_trans_handle
*trans
;
718 trans
= start_transaction(root
, 0, TRANS_ATTACH
,
719 BTRFS_RESERVE_NO_FLUSH
, true);
720 if (IS_ERR(trans
) && PTR_ERR(trans
) == -ENOENT
)
721 btrfs_wait_for_commit(root
->fs_info
, 0);
726 /* wait for a transaction commit to be fully complete */
727 static noinline
void wait_for_commit(struct btrfs_transaction
*commit
)
729 wait_event(commit
->commit_wait
, commit
->state
== TRANS_STATE_COMPLETED
);
732 int btrfs_wait_for_commit(struct btrfs_fs_info
*fs_info
, u64 transid
)
734 struct btrfs_transaction
*cur_trans
= NULL
, *t
;
738 if (transid
<= fs_info
->last_trans_committed
)
741 /* find specified transaction */
742 spin_lock(&fs_info
->trans_lock
);
743 list_for_each_entry(t
, &fs_info
->trans_list
, list
) {
744 if (t
->transid
== transid
) {
746 refcount_inc(&cur_trans
->use_count
);
750 if (t
->transid
> transid
) {
755 spin_unlock(&fs_info
->trans_lock
);
758 * The specified transaction doesn't exist, or we
759 * raced with btrfs_commit_transaction
762 if (transid
> fs_info
->last_trans_committed
)
767 /* find newest transaction that is committing | committed */
768 spin_lock(&fs_info
->trans_lock
);
769 list_for_each_entry_reverse(t
, &fs_info
->trans_list
,
771 if (t
->state
>= TRANS_STATE_COMMIT_START
) {
772 if (t
->state
== TRANS_STATE_COMPLETED
)
775 refcount_inc(&cur_trans
->use_count
);
779 spin_unlock(&fs_info
->trans_lock
);
781 goto out
; /* nothing committing|committed */
784 wait_for_commit(cur_trans
);
785 btrfs_put_transaction(cur_trans
);
790 void btrfs_throttle(struct btrfs_fs_info
*fs_info
)
792 if (!atomic_read(&fs_info
->open_ioctl_trans
))
793 wait_current_trans(fs_info
);
796 static int should_end_transaction(struct btrfs_trans_handle
*trans
)
798 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
800 if (fs_info
->global_block_rsv
.space_info
->full
&&
801 btrfs_check_space_for_delayed_refs(trans
, fs_info
))
804 return !!btrfs_block_rsv_check(&fs_info
->global_block_rsv
, 5);
807 int btrfs_should_end_transaction(struct btrfs_trans_handle
*trans
)
809 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
810 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
815 if (cur_trans
->state
>= TRANS_STATE_BLOCKED
||
816 cur_trans
->delayed_refs
.flushing
)
819 updates
= trans
->delayed_ref_updates
;
820 trans
->delayed_ref_updates
= 0;
822 err
= btrfs_run_delayed_refs(trans
, fs_info
, updates
* 2);
823 if (err
) /* Error code will also eval true */
827 return should_end_transaction(trans
);
830 static int __btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
833 struct btrfs_fs_info
*info
= trans
->fs_info
;
834 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
835 u64 transid
= trans
->transid
;
836 unsigned long cur
= trans
->delayed_ref_updates
;
837 int lock
= (trans
->type
!= TRANS_JOIN_NOLOCK
);
839 int must_run_delayed_refs
= 0;
841 if (trans
->use_count
> 1) {
843 trans
->block_rsv
= trans
->orig_rsv
;
847 btrfs_trans_release_metadata(trans
, info
);
848 trans
->block_rsv
= NULL
;
850 if (!list_empty(&trans
->new_bgs
))
851 btrfs_create_pending_block_groups(trans
, info
);
853 trans
->delayed_ref_updates
= 0;
855 must_run_delayed_refs
=
856 btrfs_should_throttle_delayed_refs(trans
, info
);
857 cur
= max_t(unsigned long, cur
, 32);
860 * don't make the caller wait if they are from a NOLOCK
861 * or ATTACH transaction, it will deadlock with commit
863 if (must_run_delayed_refs
== 1 &&
864 (trans
->type
& (__TRANS_JOIN_NOLOCK
| __TRANS_ATTACH
)))
865 must_run_delayed_refs
= 2;
868 btrfs_trans_release_metadata(trans
, info
);
869 trans
->block_rsv
= NULL
;
871 if (!list_empty(&trans
->new_bgs
))
872 btrfs_create_pending_block_groups(trans
, info
);
874 btrfs_trans_release_chunk_metadata(trans
);
876 if (lock
&& !atomic_read(&info
->open_ioctl_trans
) &&
877 should_end_transaction(trans
) &&
878 READ_ONCE(cur_trans
->state
) == TRANS_STATE_RUNNING
) {
879 spin_lock(&info
->trans_lock
);
880 if (cur_trans
->state
== TRANS_STATE_RUNNING
)
881 cur_trans
->state
= TRANS_STATE_BLOCKED
;
882 spin_unlock(&info
->trans_lock
);
885 if (lock
&& READ_ONCE(cur_trans
->state
) == TRANS_STATE_BLOCKED
) {
887 return btrfs_commit_transaction(trans
);
889 wake_up_process(info
->transaction_kthread
);
892 if (trans
->type
& __TRANS_FREEZABLE
)
893 sb_end_intwrite(info
->sb
);
895 WARN_ON(cur_trans
!= info
->running_transaction
);
896 WARN_ON(atomic_read(&cur_trans
->num_writers
) < 1);
897 atomic_dec(&cur_trans
->num_writers
);
898 extwriter_counter_dec(cur_trans
, trans
->type
);
901 * Make sure counter is updated before we wake up waiters.
904 if (waitqueue_active(&cur_trans
->writer_wait
))
905 wake_up(&cur_trans
->writer_wait
);
906 btrfs_put_transaction(cur_trans
);
908 if (current
->journal_info
== trans
)
909 current
->journal_info
= NULL
;
912 btrfs_run_delayed_iputs(info
);
914 if (trans
->aborted
||
915 test_bit(BTRFS_FS_STATE_ERROR
, &info
->fs_state
)) {
916 wake_up_process(info
->transaction_kthread
);
920 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
921 if (must_run_delayed_refs
) {
922 btrfs_async_run_delayed_refs(info
, cur
, transid
,
923 must_run_delayed_refs
== 1);
928 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
)
930 return __btrfs_end_transaction(trans
, 0);
933 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
)
935 return __btrfs_end_transaction(trans
, 1);
939 * when btree blocks are allocated, they have some corresponding bits set for
940 * them in one of two extent_io trees. This is used to make sure all of
941 * those extents are sent to disk but does not wait on them
943 int btrfs_write_marked_extents(struct btrfs_fs_info
*fs_info
,
944 struct extent_io_tree
*dirty_pages
, int mark
)
948 struct address_space
*mapping
= fs_info
->btree_inode
->i_mapping
;
949 struct extent_state
*cached_state
= NULL
;
953 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
954 mark
, &cached_state
)) {
955 bool wait_writeback
= false;
957 err
= convert_extent_bit(dirty_pages
, start
, end
,
959 mark
, &cached_state
);
961 * convert_extent_bit can return -ENOMEM, which is most of the
962 * time a temporary error. So when it happens, ignore the error
963 * and wait for writeback of this range to finish - because we
964 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
965 * to __btrfs_wait_marked_extents() would not know that
966 * writeback for this range started and therefore wouldn't
967 * wait for it to finish - we don't want to commit a
968 * superblock that points to btree nodes/leafs for which
969 * writeback hasn't finished yet (and without errors).
970 * We cleanup any entries left in the io tree when committing
971 * the transaction (through clear_btree_io_tree()).
973 if (err
== -ENOMEM
) {
975 wait_writeback
= true;
978 err
= filemap_fdatawrite_range(mapping
, start
, end
);
981 else if (wait_writeback
)
982 werr
= filemap_fdatawait_range(mapping
, start
, end
);
983 free_extent_state(cached_state
);
992 * when btree blocks are allocated, they have some corresponding bits set for
993 * them in one of two extent_io trees. This is used to make sure all of
994 * those extents are on disk for transaction or log commit. We wait
995 * on all the pages and clear them from the dirty pages state tree
997 static int __btrfs_wait_marked_extents(struct btrfs_fs_info
*fs_info
,
998 struct extent_io_tree
*dirty_pages
)
1002 struct address_space
*mapping
= fs_info
->btree_inode
->i_mapping
;
1003 struct extent_state
*cached_state
= NULL
;
1007 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
1008 EXTENT_NEED_WAIT
, &cached_state
)) {
1010 * Ignore -ENOMEM errors returned by clear_extent_bit().
1011 * When committing the transaction, we'll remove any entries
1012 * left in the io tree. For a log commit, we don't remove them
1013 * after committing the log because the tree can be accessed
1014 * concurrently - we do it only at transaction commit time when
1015 * it's safe to do it (through clear_btree_io_tree()).
1017 err
= clear_extent_bit(dirty_pages
, start
, end
,
1019 0, 0, &cached_state
, GFP_NOFS
);
1023 err
= filemap_fdatawait_range(mapping
, start
, end
);
1026 free_extent_state(cached_state
);
1027 cached_state
= NULL
;
1036 int btrfs_wait_extents(struct btrfs_fs_info
*fs_info
,
1037 struct extent_io_tree
*dirty_pages
)
1039 bool errors
= false;
1042 err
= __btrfs_wait_marked_extents(fs_info
, dirty_pages
);
1043 if (test_and_clear_bit(BTRFS_FS_BTREE_ERR
, &fs_info
->flags
))
1051 int btrfs_wait_tree_log_extents(struct btrfs_root
*log_root
, int mark
)
1053 struct btrfs_fs_info
*fs_info
= log_root
->fs_info
;
1054 struct extent_io_tree
*dirty_pages
= &log_root
->dirty_log_pages
;
1055 bool errors
= false;
1058 ASSERT(log_root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
);
1060 err
= __btrfs_wait_marked_extents(fs_info
, dirty_pages
);
1061 if ((mark
& EXTENT_DIRTY
) &&
1062 test_and_clear_bit(BTRFS_FS_LOG1_ERR
, &fs_info
->flags
))
1065 if ((mark
& EXTENT_NEW
) &&
1066 test_and_clear_bit(BTRFS_FS_LOG2_ERR
, &fs_info
->flags
))
1075 * when btree blocks are allocated, they have some corresponding bits set for
1076 * them in one of two extent_io trees. This is used to make sure all of
1077 * those extents are on disk for transaction or log commit
1079 static int btrfs_write_and_wait_marked_extents(struct btrfs_fs_info
*fs_info
,
1080 struct extent_io_tree
*dirty_pages
, int mark
)
1084 struct blk_plug plug
;
1086 blk_start_plug(&plug
);
1087 ret
= btrfs_write_marked_extents(fs_info
, dirty_pages
, mark
);
1088 blk_finish_plug(&plug
);
1089 ret2
= btrfs_wait_extents(fs_info
, dirty_pages
);
1098 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
,
1099 struct btrfs_fs_info
*fs_info
)
1103 ret
= btrfs_write_and_wait_marked_extents(fs_info
,
1104 &trans
->transaction
->dirty_pages
,
1106 clear_btree_io_tree(&trans
->transaction
->dirty_pages
);
1112 * this is used to update the root pointer in the tree of tree roots.
1114 * But, in the case of the extent allocation tree, updating the root
1115 * pointer may allocate blocks which may change the root of the extent
1118 * So, this loops and repeats and makes sure the cowonly root didn't
1119 * change while the root pointer was being updated in the metadata.
1121 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
1122 struct btrfs_root
*root
)
1125 u64 old_root_bytenr
;
1127 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1128 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1130 old_root_used
= btrfs_root_used(&root
->root_item
);
1133 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
1134 if (old_root_bytenr
== root
->node
->start
&&
1135 old_root_used
== btrfs_root_used(&root
->root_item
))
1138 btrfs_set_root_node(&root
->root_item
, root
->node
);
1139 ret
= btrfs_update_root(trans
, tree_root
,
1145 old_root_used
= btrfs_root_used(&root
->root_item
);
1152 * update all the cowonly tree roots on disk
1154 * The error handling in this function may not be obvious. Any of the
1155 * failures will cause the file system to go offline. We still need
1156 * to clean up the delayed refs.
1158 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
,
1159 struct btrfs_fs_info
*fs_info
)
1161 struct list_head
*dirty_bgs
= &trans
->transaction
->dirty_bgs
;
1162 struct list_head
*io_bgs
= &trans
->transaction
->io_bgs
;
1163 struct list_head
*next
;
1164 struct extent_buffer
*eb
;
1167 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
1168 ret
= btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
,
1170 btrfs_tree_unlock(eb
);
1171 free_extent_buffer(eb
);
1176 ret
= btrfs_run_delayed_refs(trans
, fs_info
, (unsigned long)-1);
1180 ret
= btrfs_run_dev_stats(trans
, fs_info
);
1183 ret
= btrfs_run_dev_replace(trans
, fs_info
);
1186 ret
= btrfs_run_qgroups(trans
, fs_info
);
1190 ret
= btrfs_setup_space_cache(trans
, fs_info
);
1194 /* run_qgroups might have added some more refs */
1195 ret
= btrfs_run_delayed_refs(trans
, fs_info
, (unsigned long)-1);
1199 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
1200 struct btrfs_root
*root
;
1201 next
= fs_info
->dirty_cowonly_roots
.next
;
1202 list_del_init(next
);
1203 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
1204 clear_bit(BTRFS_ROOT_DIRTY
, &root
->state
);
1206 if (root
!= fs_info
->extent_root
)
1207 list_add_tail(&root
->dirty_list
,
1208 &trans
->transaction
->switch_commits
);
1209 ret
= update_cowonly_root(trans
, root
);
1212 ret
= btrfs_run_delayed_refs(trans
, fs_info
, (unsigned long)-1);
1217 while (!list_empty(dirty_bgs
) || !list_empty(io_bgs
)) {
1218 ret
= btrfs_write_dirty_block_groups(trans
, fs_info
);
1221 ret
= btrfs_run_delayed_refs(trans
, fs_info
, (unsigned long)-1);
1226 if (!list_empty(&fs_info
->dirty_cowonly_roots
))
1229 list_add_tail(&fs_info
->extent_root
->dirty_list
,
1230 &trans
->transaction
->switch_commits
);
1231 btrfs_after_dev_replace_commit(fs_info
);
1237 * dead roots are old snapshots that need to be deleted. This allocates
1238 * a dirty root struct and adds it into the list of dead roots that need to
1241 void btrfs_add_dead_root(struct btrfs_root
*root
)
1243 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1245 spin_lock(&fs_info
->trans_lock
);
1246 if (list_empty(&root
->root_list
))
1247 list_add_tail(&root
->root_list
, &fs_info
->dead_roots
);
1248 spin_unlock(&fs_info
->trans_lock
);
1252 * update all the cowonly tree roots on disk
1254 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
,
1255 struct btrfs_fs_info
*fs_info
)
1257 struct btrfs_root
*gang
[8];
1262 spin_lock(&fs_info
->fs_roots_radix_lock
);
1264 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
1267 BTRFS_ROOT_TRANS_TAG
);
1270 for (i
= 0; i
< ret
; i
++) {
1271 struct btrfs_root
*root
= gang
[i
];
1272 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
1273 (unsigned long)root
->root_key
.objectid
,
1274 BTRFS_ROOT_TRANS_TAG
);
1275 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1277 btrfs_free_log(trans
, root
);
1278 btrfs_update_reloc_root(trans
, root
);
1279 btrfs_orphan_commit_root(trans
, root
);
1281 btrfs_save_ino_cache(root
, trans
);
1283 /* see comments in should_cow_block() */
1284 clear_bit(BTRFS_ROOT_FORCE_COW
, &root
->state
);
1285 smp_mb__after_atomic();
1287 if (root
->commit_root
!= root
->node
) {
1288 list_add_tail(&root
->dirty_list
,
1289 &trans
->transaction
->switch_commits
);
1290 btrfs_set_root_node(&root
->root_item
,
1294 err
= btrfs_update_root(trans
, fs_info
->tree_root
,
1297 spin_lock(&fs_info
->fs_roots_radix_lock
);
1300 btrfs_qgroup_free_meta_all(root
);
1303 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1308 * defrag a given btree.
1309 * Every leaf in the btree is read and defragged.
1311 int btrfs_defrag_root(struct btrfs_root
*root
)
1313 struct btrfs_fs_info
*info
= root
->fs_info
;
1314 struct btrfs_trans_handle
*trans
;
1317 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING
, &root
->state
))
1321 trans
= btrfs_start_transaction(root
, 0);
1323 return PTR_ERR(trans
);
1325 ret
= btrfs_defrag_leaves(trans
, root
);
1327 btrfs_end_transaction(trans
);
1328 btrfs_btree_balance_dirty(info
);
1331 if (btrfs_fs_closing(info
) || ret
!= -EAGAIN
)
1334 if (btrfs_defrag_cancelled(info
)) {
1335 btrfs_debug(info
, "defrag_root cancelled");
1340 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING
, &root
->state
);
1345 * Do all special snapshot related qgroup dirty hack.
1347 * Will do all needed qgroup inherit and dirty hack like switch commit
1348 * roots inside one transaction and write all btree into disk, to make
1351 static int qgroup_account_snapshot(struct btrfs_trans_handle
*trans
,
1352 struct btrfs_root
*src
,
1353 struct btrfs_root
*parent
,
1354 struct btrfs_qgroup_inherit
*inherit
,
1357 struct btrfs_fs_info
*fs_info
= src
->fs_info
;
1361 * Save some performance in the case that qgroups are not
1362 * enabled. If this check races with the ioctl, rescan will
1365 if (!test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
))
1369 * Ensure dirty @src will be commited. Or, after comming
1370 * commit_fs_roots() and switch_commit_roots(), any dirty but not
1371 * recorded root will never be updated again, causing an outdated root
1374 record_root_in_trans(trans
, src
, 1);
1377 * We are going to commit transaction, see btrfs_commit_transaction()
1378 * comment for reason locking tree_log_mutex
1380 mutex_lock(&fs_info
->tree_log_mutex
);
1382 ret
= commit_fs_roots(trans
, fs_info
);
1385 ret
= btrfs_qgroup_account_extents(trans
, fs_info
);
1389 /* Now qgroup are all updated, we can inherit it to new qgroups */
1390 ret
= btrfs_qgroup_inherit(trans
, fs_info
,
1391 src
->root_key
.objectid
, dst_objectid
,
1397 * Now we do a simplified commit transaction, which will:
1398 * 1) commit all subvolume and extent tree
1399 * To ensure all subvolume and extent tree have a valid
1400 * commit_root to accounting later insert_dir_item()
1401 * 2) write all btree blocks onto disk
1402 * This is to make sure later btree modification will be cowed
1403 * Or commit_root can be populated and cause wrong qgroup numbers
1404 * In this simplified commit, we don't really care about other trees
1405 * like chunk and root tree, as they won't affect qgroup.
1406 * And we don't write super to avoid half committed status.
1408 ret
= commit_cowonly_roots(trans
, fs_info
);
1411 switch_commit_roots(trans
->transaction
, fs_info
);
1412 ret
= btrfs_write_and_wait_transaction(trans
, fs_info
);
1414 btrfs_handle_fs_error(fs_info
, ret
,
1415 "Error while writing out transaction for qgroup");
1418 mutex_unlock(&fs_info
->tree_log_mutex
);
1421 * Force parent root to be updated, as we recorded it before so its
1422 * last_trans == cur_transid.
1423 * Or it won't be committed again onto disk after later
1427 record_root_in_trans(trans
, parent
, 1);
1432 * new snapshots need to be created at a very specific time in the
1433 * transaction commit. This does the actual creation.
1436 * If the error which may affect the commitment of the current transaction
1437 * happens, we should return the error number. If the error which just affect
1438 * the creation of the pending snapshots, just return 0.
1440 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
1441 struct btrfs_fs_info
*fs_info
,
1442 struct btrfs_pending_snapshot
*pending
)
1444 struct btrfs_key key
;
1445 struct btrfs_root_item
*new_root_item
;
1446 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1447 struct btrfs_root
*root
= pending
->root
;
1448 struct btrfs_root
*parent_root
;
1449 struct btrfs_block_rsv
*rsv
;
1450 struct inode
*parent_inode
;
1451 struct btrfs_path
*path
;
1452 struct btrfs_dir_item
*dir_item
;
1453 struct dentry
*dentry
;
1454 struct extent_buffer
*tmp
;
1455 struct extent_buffer
*old
;
1456 struct timespec cur_time
;
1464 ASSERT(pending
->path
);
1465 path
= pending
->path
;
1467 ASSERT(pending
->root_item
);
1468 new_root_item
= pending
->root_item
;
1470 pending
->error
= btrfs_find_free_objectid(tree_root
, &objectid
);
1472 goto no_free_objectid
;
1475 * Make qgroup to skip current new snapshot's qgroupid, as it is
1476 * accounted by later btrfs_qgroup_inherit().
1478 btrfs_set_skip_qgroup(trans
, objectid
);
1480 btrfs_reloc_pre_snapshot(pending
, &to_reserve
);
1482 if (to_reserve
> 0) {
1483 pending
->error
= btrfs_block_rsv_add(root
,
1484 &pending
->block_rsv
,
1486 BTRFS_RESERVE_NO_FLUSH
);
1488 goto clear_skip_qgroup
;
1491 key
.objectid
= objectid
;
1492 key
.offset
= (u64
)-1;
1493 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1495 rsv
= trans
->block_rsv
;
1496 trans
->block_rsv
= &pending
->block_rsv
;
1497 trans
->bytes_reserved
= trans
->block_rsv
->reserved
;
1498 trace_btrfs_space_reservation(fs_info
, "transaction",
1500 trans
->bytes_reserved
, 1);
1501 dentry
= pending
->dentry
;
1502 parent_inode
= pending
->dir
;
1503 parent_root
= BTRFS_I(parent_inode
)->root
;
1504 record_root_in_trans(trans
, parent_root
, 0);
1506 cur_time
= current_time(parent_inode
);
1509 * insert the directory item
1511 ret
= btrfs_set_inode_index(BTRFS_I(parent_inode
), &index
);
1512 BUG_ON(ret
); /* -ENOMEM */
1514 /* check if there is a file/dir which has the same name. */
1515 dir_item
= btrfs_lookup_dir_item(NULL
, parent_root
, path
,
1516 btrfs_ino(BTRFS_I(parent_inode
)),
1517 dentry
->d_name
.name
,
1518 dentry
->d_name
.len
, 0);
1519 if (dir_item
!= NULL
&& !IS_ERR(dir_item
)) {
1520 pending
->error
= -EEXIST
;
1521 goto dir_item_existed
;
1522 } else if (IS_ERR(dir_item
)) {
1523 ret
= PTR_ERR(dir_item
);
1524 btrfs_abort_transaction(trans
, ret
);
1527 btrfs_release_path(path
);
1530 * pull in the delayed directory update
1531 * and the delayed inode item
1532 * otherwise we corrupt the FS during
1535 ret
= btrfs_run_delayed_items(trans
, fs_info
);
1536 if (ret
) { /* Transaction aborted */
1537 btrfs_abort_transaction(trans
, ret
);
1541 record_root_in_trans(trans
, root
, 0);
1542 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
1543 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
1544 btrfs_check_and_init_root_item(new_root_item
);
1546 root_flags
= btrfs_root_flags(new_root_item
);
1547 if (pending
->readonly
)
1548 root_flags
|= BTRFS_ROOT_SUBVOL_RDONLY
;
1550 root_flags
&= ~BTRFS_ROOT_SUBVOL_RDONLY
;
1551 btrfs_set_root_flags(new_root_item
, root_flags
);
1553 btrfs_set_root_generation_v2(new_root_item
,
1555 uuid_le_gen(&new_uuid
);
1556 memcpy(new_root_item
->uuid
, new_uuid
.b
, BTRFS_UUID_SIZE
);
1557 memcpy(new_root_item
->parent_uuid
, root
->root_item
.uuid
,
1559 if (!(root_flags
& BTRFS_ROOT_SUBVOL_RDONLY
)) {
1560 memset(new_root_item
->received_uuid
, 0,
1561 sizeof(new_root_item
->received_uuid
));
1562 memset(&new_root_item
->stime
, 0, sizeof(new_root_item
->stime
));
1563 memset(&new_root_item
->rtime
, 0, sizeof(new_root_item
->rtime
));
1564 btrfs_set_root_stransid(new_root_item
, 0);
1565 btrfs_set_root_rtransid(new_root_item
, 0);
1567 btrfs_set_stack_timespec_sec(&new_root_item
->otime
, cur_time
.tv_sec
);
1568 btrfs_set_stack_timespec_nsec(&new_root_item
->otime
, cur_time
.tv_nsec
);
1569 btrfs_set_root_otransid(new_root_item
, trans
->transid
);
1571 old
= btrfs_lock_root_node(root
);
1572 ret
= btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
);
1574 btrfs_tree_unlock(old
);
1575 free_extent_buffer(old
);
1576 btrfs_abort_transaction(trans
, ret
);
1580 btrfs_set_lock_blocking(old
);
1582 ret
= btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
1583 /* clean up in any case */
1584 btrfs_tree_unlock(old
);
1585 free_extent_buffer(old
);
1587 btrfs_abort_transaction(trans
, ret
);
1590 /* see comments in should_cow_block() */
1591 set_bit(BTRFS_ROOT_FORCE_COW
, &root
->state
);
1594 btrfs_set_root_node(new_root_item
, tmp
);
1595 /* record when the snapshot was created in key.offset */
1596 key
.offset
= trans
->transid
;
1597 ret
= btrfs_insert_root(trans
, tree_root
, &key
, new_root_item
);
1598 btrfs_tree_unlock(tmp
);
1599 free_extent_buffer(tmp
);
1601 btrfs_abort_transaction(trans
, ret
);
1606 * insert root back/forward references
1608 ret
= btrfs_add_root_ref(trans
, fs_info
, objectid
,
1609 parent_root
->root_key
.objectid
,
1610 btrfs_ino(BTRFS_I(parent_inode
)), index
,
1611 dentry
->d_name
.name
, dentry
->d_name
.len
);
1613 btrfs_abort_transaction(trans
, ret
);
1617 key
.offset
= (u64
)-1;
1618 pending
->snap
= btrfs_read_fs_root_no_name(fs_info
, &key
);
1619 if (IS_ERR(pending
->snap
)) {
1620 ret
= PTR_ERR(pending
->snap
);
1621 btrfs_abort_transaction(trans
, ret
);
1625 ret
= btrfs_reloc_post_snapshot(trans
, pending
);
1627 btrfs_abort_transaction(trans
, ret
);
1631 ret
= btrfs_run_delayed_refs(trans
, fs_info
, (unsigned long)-1);
1633 btrfs_abort_transaction(trans
, ret
);
1638 * Do special qgroup accounting for snapshot, as we do some qgroup
1639 * snapshot hack to do fast snapshot.
1640 * To co-operate with that hack, we do hack again.
1641 * Or snapshot will be greatly slowed down by a subtree qgroup rescan
1643 ret
= qgroup_account_snapshot(trans
, root
, parent_root
,
1644 pending
->inherit
, objectid
);
1648 ret
= btrfs_insert_dir_item(trans
, parent_root
,
1649 dentry
->d_name
.name
, dentry
->d_name
.len
,
1650 BTRFS_I(parent_inode
), &key
,
1651 BTRFS_FT_DIR
, index
);
1652 /* We have check then name at the beginning, so it is impossible. */
1653 BUG_ON(ret
== -EEXIST
|| ret
== -EOVERFLOW
);
1655 btrfs_abort_transaction(trans
, ret
);
1659 btrfs_i_size_write(BTRFS_I(parent_inode
), parent_inode
->i_size
+
1660 dentry
->d_name
.len
* 2);
1661 parent_inode
->i_mtime
= parent_inode
->i_ctime
=
1662 current_time(parent_inode
);
1663 ret
= btrfs_update_inode_fallback(trans
, parent_root
, parent_inode
);
1665 btrfs_abort_transaction(trans
, ret
);
1668 ret
= btrfs_uuid_tree_add(trans
, fs_info
, new_uuid
.b
,
1669 BTRFS_UUID_KEY_SUBVOL
, objectid
);
1671 btrfs_abort_transaction(trans
, ret
);
1674 if (!btrfs_is_empty_uuid(new_root_item
->received_uuid
)) {
1675 ret
= btrfs_uuid_tree_add(trans
, fs_info
,
1676 new_root_item
->received_uuid
,
1677 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
1679 if (ret
&& ret
!= -EEXIST
) {
1680 btrfs_abort_transaction(trans
, ret
);
1685 ret
= btrfs_run_delayed_refs(trans
, fs_info
, (unsigned long)-1);
1687 btrfs_abort_transaction(trans
, ret
);
1692 pending
->error
= ret
;
1694 trans
->block_rsv
= rsv
;
1695 trans
->bytes_reserved
= 0;
1697 btrfs_clear_skip_qgroup(trans
);
1699 kfree(new_root_item
);
1700 pending
->root_item
= NULL
;
1701 btrfs_free_path(path
);
1702 pending
->path
= NULL
;
1708 * create all the snapshots we've scheduled for creation
1710 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
,
1711 struct btrfs_fs_info
*fs_info
)
1713 struct btrfs_pending_snapshot
*pending
, *next
;
1714 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
1717 list_for_each_entry_safe(pending
, next
, head
, list
) {
1718 list_del(&pending
->list
);
1719 ret
= create_pending_snapshot(trans
, fs_info
, pending
);
1726 static void update_super_roots(struct btrfs_fs_info
*fs_info
)
1728 struct btrfs_root_item
*root_item
;
1729 struct btrfs_super_block
*super
;
1731 super
= fs_info
->super_copy
;
1733 root_item
= &fs_info
->chunk_root
->root_item
;
1734 super
->chunk_root
= root_item
->bytenr
;
1735 super
->chunk_root_generation
= root_item
->generation
;
1736 super
->chunk_root_level
= root_item
->level
;
1738 root_item
= &fs_info
->tree_root
->root_item
;
1739 super
->root
= root_item
->bytenr
;
1740 super
->generation
= root_item
->generation
;
1741 super
->root_level
= root_item
->level
;
1742 if (btrfs_test_opt(fs_info
, SPACE_CACHE
))
1743 super
->cache_generation
= root_item
->generation
;
1744 if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN
, &fs_info
->flags
))
1745 super
->uuid_tree_generation
= root_item
->generation
;
1748 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
1750 struct btrfs_transaction
*trans
;
1753 spin_lock(&info
->trans_lock
);
1754 trans
= info
->running_transaction
;
1756 ret
= (trans
->state
>= TRANS_STATE_COMMIT_START
);
1757 spin_unlock(&info
->trans_lock
);
1761 int btrfs_transaction_blocked(struct btrfs_fs_info
*info
)
1763 struct btrfs_transaction
*trans
;
1766 spin_lock(&info
->trans_lock
);
1767 trans
= info
->running_transaction
;
1769 ret
= is_transaction_blocked(trans
);
1770 spin_unlock(&info
->trans_lock
);
1775 * wait for the current transaction commit to start and block subsequent
1778 static void wait_current_trans_commit_start(struct btrfs_fs_info
*fs_info
,
1779 struct btrfs_transaction
*trans
)
1781 wait_event(fs_info
->transaction_blocked_wait
,
1782 trans
->state
>= TRANS_STATE_COMMIT_START
|| trans
->aborted
);
1786 * wait for the current transaction to start and then become unblocked.
1789 static void wait_current_trans_commit_start_and_unblock(
1790 struct btrfs_fs_info
*fs_info
,
1791 struct btrfs_transaction
*trans
)
1793 wait_event(fs_info
->transaction_wait
,
1794 trans
->state
>= TRANS_STATE_UNBLOCKED
|| trans
->aborted
);
1798 * commit transactions asynchronously. once btrfs_commit_transaction_async
1799 * returns, any subsequent transaction will not be allowed to join.
1801 struct btrfs_async_commit
{
1802 struct btrfs_trans_handle
*newtrans
;
1803 struct work_struct work
;
1806 static void do_async_commit(struct work_struct
*work
)
1808 struct btrfs_async_commit
*ac
=
1809 container_of(work
, struct btrfs_async_commit
, work
);
1812 * We've got freeze protection passed with the transaction.
1813 * Tell lockdep about it.
1815 if (ac
->newtrans
->type
& __TRANS_FREEZABLE
)
1816 __sb_writers_acquired(ac
->newtrans
->fs_info
->sb
, SB_FREEZE_FS
);
1818 current
->journal_info
= ac
->newtrans
;
1820 btrfs_commit_transaction(ac
->newtrans
);
1824 int btrfs_commit_transaction_async(struct btrfs_trans_handle
*trans
,
1825 int wait_for_unblock
)
1827 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1828 struct btrfs_async_commit
*ac
;
1829 struct btrfs_transaction
*cur_trans
;
1831 ac
= kmalloc(sizeof(*ac
), GFP_NOFS
);
1835 INIT_WORK(&ac
->work
, do_async_commit
);
1836 ac
->newtrans
= btrfs_join_transaction(trans
->root
);
1837 if (IS_ERR(ac
->newtrans
)) {
1838 int err
= PTR_ERR(ac
->newtrans
);
1843 /* take transaction reference */
1844 cur_trans
= trans
->transaction
;
1845 refcount_inc(&cur_trans
->use_count
);
1847 btrfs_end_transaction(trans
);
1850 * Tell lockdep we've released the freeze rwsem, since the
1851 * async commit thread will be the one to unlock it.
1853 if (ac
->newtrans
->type
& __TRANS_FREEZABLE
)
1854 __sb_writers_release(fs_info
->sb
, SB_FREEZE_FS
);
1856 schedule_work(&ac
->work
);
1858 /* wait for transaction to start and unblock */
1859 if (wait_for_unblock
)
1860 wait_current_trans_commit_start_and_unblock(fs_info
, cur_trans
);
1862 wait_current_trans_commit_start(fs_info
, cur_trans
);
1864 if (current
->journal_info
== trans
)
1865 current
->journal_info
= NULL
;
1867 btrfs_put_transaction(cur_trans
);
1872 static void cleanup_transaction(struct btrfs_trans_handle
*trans
,
1873 struct btrfs_root
*root
, int err
)
1875 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1876 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1879 WARN_ON(trans
->use_count
> 1);
1881 btrfs_abort_transaction(trans
, err
);
1883 spin_lock(&fs_info
->trans_lock
);
1886 * If the transaction is removed from the list, it means this
1887 * transaction has been committed successfully, so it is impossible
1888 * to call the cleanup function.
1890 BUG_ON(list_empty(&cur_trans
->list
));
1892 list_del_init(&cur_trans
->list
);
1893 if (cur_trans
== fs_info
->running_transaction
) {
1894 cur_trans
->state
= TRANS_STATE_COMMIT_DOING
;
1895 spin_unlock(&fs_info
->trans_lock
);
1896 wait_event(cur_trans
->writer_wait
,
1897 atomic_read(&cur_trans
->num_writers
) == 1);
1899 spin_lock(&fs_info
->trans_lock
);
1901 spin_unlock(&fs_info
->trans_lock
);
1903 btrfs_cleanup_one_transaction(trans
->transaction
, fs_info
);
1905 spin_lock(&fs_info
->trans_lock
);
1906 if (cur_trans
== fs_info
->running_transaction
)
1907 fs_info
->running_transaction
= NULL
;
1908 spin_unlock(&fs_info
->trans_lock
);
1910 if (trans
->type
& __TRANS_FREEZABLE
)
1911 sb_end_intwrite(fs_info
->sb
);
1912 btrfs_put_transaction(cur_trans
);
1913 btrfs_put_transaction(cur_trans
);
1915 trace_btrfs_transaction_commit(root
);
1917 if (current
->journal_info
== trans
)
1918 current
->journal_info
= NULL
;
1919 btrfs_scrub_cancel(fs_info
);
1921 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1924 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info
*fs_info
)
1926 if (btrfs_test_opt(fs_info
, FLUSHONCOMMIT
))
1927 return btrfs_start_delalloc_roots(fs_info
, 1, -1);
1931 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info
*fs_info
)
1933 if (btrfs_test_opt(fs_info
, FLUSHONCOMMIT
))
1934 btrfs_wait_ordered_roots(fs_info
, U64_MAX
, 0, (u64
)-1);
1938 btrfs_wait_pending_ordered(struct btrfs_transaction
*cur_trans
)
1940 wait_event(cur_trans
->pending_wait
,
1941 atomic_read(&cur_trans
->pending_ordered
) == 0);
1944 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
)
1946 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1947 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1948 struct btrfs_transaction
*prev_trans
= NULL
;
1952 * Some places just start a transaction to commit it. We need to make
1953 * sure that if this commit fails that the abort code actually marks the
1954 * transaction as failed, so set trans->dirty to make the abort code do
1957 trans
->dirty
= true;
1959 /* Stop the commit early if ->aborted is set */
1960 if (unlikely(READ_ONCE(cur_trans
->aborted
))) {
1961 ret
= cur_trans
->aborted
;
1962 btrfs_end_transaction(trans
);
1966 btrfs_trans_release_metadata(trans
, fs_info
);
1967 trans
->block_rsv
= NULL
;
1969 /* make a pass through all the delayed refs we have so far
1970 * any runnings procs may add more while we are here
1972 ret
= btrfs_run_delayed_refs(trans
, fs_info
, 0);
1974 btrfs_end_transaction(trans
);
1978 cur_trans
= trans
->transaction
;
1981 * set the flushing flag so procs in this transaction have to
1982 * start sending their work down.
1984 cur_trans
->delayed_refs
.flushing
= 1;
1987 if (!list_empty(&trans
->new_bgs
))
1988 btrfs_create_pending_block_groups(trans
, fs_info
);
1990 ret
= btrfs_run_delayed_refs(trans
, fs_info
, 0);
1992 btrfs_end_transaction(trans
);
1996 if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN
, &cur_trans
->flags
)) {
1999 /* this mutex is also taken before trying to set
2000 * block groups readonly. We need to make sure
2001 * that nobody has set a block group readonly
2002 * after a extents from that block group have been
2003 * allocated for cache files. btrfs_set_block_group_ro
2004 * will wait for the transaction to commit if it
2005 * finds BTRFS_TRANS_DIRTY_BG_RUN set.
2007 * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure
2008 * only one process starts all the block group IO. It wouldn't
2009 * hurt to have more than one go through, but there's no
2010 * real advantage to it either.
2012 mutex_lock(&fs_info
->ro_block_group_mutex
);
2013 if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN
,
2016 mutex_unlock(&fs_info
->ro_block_group_mutex
);
2019 ret
= btrfs_start_dirty_block_groups(trans
, fs_info
);
2022 btrfs_end_transaction(trans
);
2026 spin_lock(&fs_info
->trans_lock
);
2027 if (cur_trans
->state
>= TRANS_STATE_COMMIT_START
) {
2028 spin_unlock(&fs_info
->trans_lock
);
2029 refcount_inc(&cur_trans
->use_count
);
2030 ret
= btrfs_end_transaction(trans
);
2032 wait_for_commit(cur_trans
);
2034 if (unlikely(cur_trans
->aborted
))
2035 ret
= cur_trans
->aborted
;
2037 btrfs_put_transaction(cur_trans
);
2042 cur_trans
->state
= TRANS_STATE_COMMIT_START
;
2043 wake_up(&fs_info
->transaction_blocked_wait
);
2045 if (cur_trans
->list
.prev
!= &fs_info
->trans_list
) {
2046 prev_trans
= list_entry(cur_trans
->list
.prev
,
2047 struct btrfs_transaction
, list
);
2048 if (prev_trans
->state
!= TRANS_STATE_COMPLETED
) {
2049 refcount_inc(&prev_trans
->use_count
);
2050 spin_unlock(&fs_info
->trans_lock
);
2052 wait_for_commit(prev_trans
);
2053 ret
= prev_trans
->aborted
;
2055 btrfs_put_transaction(prev_trans
);
2057 goto cleanup_transaction
;
2059 spin_unlock(&fs_info
->trans_lock
);
2062 spin_unlock(&fs_info
->trans_lock
);
2064 * The previous transaction was aborted and was already removed
2065 * from the list of transactions at fs_info->trans_list. So we
2066 * abort to prevent writing a new superblock that reflects a
2067 * corrupt state (pointing to trees with unwritten nodes/leafs).
2069 if (test_bit(BTRFS_FS_STATE_TRANS_ABORTED
, &fs_info
->fs_state
)) {
2071 goto cleanup_transaction
;
2075 extwriter_counter_dec(cur_trans
, trans
->type
);
2077 ret
= btrfs_start_delalloc_flush(fs_info
);
2079 goto cleanup_transaction
;
2081 ret
= btrfs_run_delayed_items(trans
, fs_info
);
2083 goto cleanup_transaction
;
2085 wait_event(cur_trans
->writer_wait
,
2086 extwriter_counter_read(cur_trans
) == 0);
2088 /* some pending stuffs might be added after the previous flush. */
2089 ret
= btrfs_run_delayed_items(trans
, fs_info
);
2091 goto cleanup_transaction
;
2093 btrfs_wait_delalloc_flush(fs_info
);
2095 btrfs_wait_pending_ordered(cur_trans
);
2097 btrfs_scrub_pause(fs_info
);
2099 * Ok now we need to make sure to block out any other joins while we
2100 * commit the transaction. We could have started a join before setting
2101 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
2103 spin_lock(&fs_info
->trans_lock
);
2104 cur_trans
->state
= TRANS_STATE_COMMIT_DOING
;
2105 spin_unlock(&fs_info
->trans_lock
);
2106 wait_event(cur_trans
->writer_wait
,
2107 atomic_read(&cur_trans
->num_writers
) == 1);
2109 /* ->aborted might be set after the previous check, so check it */
2110 if (unlikely(READ_ONCE(cur_trans
->aborted
))) {
2111 ret
= cur_trans
->aborted
;
2112 goto scrub_continue
;
2115 * the reloc mutex makes sure that we stop
2116 * the balancing code from coming in and moving
2117 * extents around in the middle of the commit
2119 mutex_lock(&fs_info
->reloc_mutex
);
2122 * We needn't worry about the delayed items because we will
2123 * deal with them in create_pending_snapshot(), which is the
2124 * core function of the snapshot creation.
2126 ret
= create_pending_snapshots(trans
, fs_info
);
2128 mutex_unlock(&fs_info
->reloc_mutex
);
2129 goto scrub_continue
;
2133 * We insert the dir indexes of the snapshots and update the inode
2134 * of the snapshots' parents after the snapshot creation, so there
2135 * are some delayed items which are not dealt with. Now deal with
2138 * We needn't worry that this operation will corrupt the snapshots,
2139 * because all the tree which are snapshoted will be forced to COW
2140 * the nodes and leaves.
2142 ret
= btrfs_run_delayed_items(trans
, fs_info
);
2144 mutex_unlock(&fs_info
->reloc_mutex
);
2145 goto scrub_continue
;
2148 ret
= btrfs_run_delayed_refs(trans
, fs_info
, (unsigned long)-1);
2150 mutex_unlock(&fs_info
->reloc_mutex
);
2151 goto scrub_continue
;
2155 * make sure none of the code above managed to slip in a
2158 btrfs_assert_delayed_root_empty(fs_info
);
2160 WARN_ON(cur_trans
!= trans
->transaction
);
2162 /* btrfs_commit_tree_roots is responsible for getting the
2163 * various roots consistent with each other. Every pointer
2164 * in the tree of tree roots has to point to the most up to date
2165 * root for every subvolume and other tree. So, we have to keep
2166 * the tree logging code from jumping in and changing any
2169 * At this point in the commit, there can't be any tree-log
2170 * writers, but a little lower down we drop the trans mutex
2171 * and let new people in. By holding the tree_log_mutex
2172 * from now until after the super is written, we avoid races
2173 * with the tree-log code.
2175 mutex_lock(&fs_info
->tree_log_mutex
);
2177 ret
= commit_fs_roots(trans
, fs_info
);
2179 mutex_unlock(&fs_info
->tree_log_mutex
);
2180 mutex_unlock(&fs_info
->reloc_mutex
);
2181 goto scrub_continue
;
2185 * Since the transaction is done, we can apply the pending changes
2186 * before the next transaction.
2188 btrfs_apply_pending_changes(fs_info
);
2190 /* commit_fs_roots gets rid of all the tree log roots, it is now
2191 * safe to free the root of tree log roots
2193 btrfs_free_log_root_tree(trans
, fs_info
);
2196 * commit_fs_roots() can call btrfs_save_ino_cache(), which generates
2197 * new delayed refs. Must handle them or qgroup can be wrong.
2199 ret
= btrfs_run_delayed_refs(trans
, fs_info
, (unsigned long)-1);
2201 mutex_unlock(&fs_info
->tree_log_mutex
);
2202 mutex_unlock(&fs_info
->reloc_mutex
);
2203 goto scrub_continue
;
2207 * Since fs roots are all committed, we can get a quite accurate
2208 * new_roots. So let's do quota accounting.
2210 ret
= btrfs_qgroup_account_extents(trans
, fs_info
);
2212 mutex_unlock(&fs_info
->tree_log_mutex
);
2213 mutex_unlock(&fs_info
->reloc_mutex
);
2214 goto scrub_continue
;
2217 ret
= commit_cowonly_roots(trans
, fs_info
);
2219 mutex_unlock(&fs_info
->tree_log_mutex
);
2220 mutex_unlock(&fs_info
->reloc_mutex
);
2221 goto scrub_continue
;
2225 * The tasks which save the space cache and inode cache may also
2226 * update ->aborted, check it.
2228 if (unlikely(READ_ONCE(cur_trans
->aborted
))) {
2229 ret
= cur_trans
->aborted
;
2230 mutex_unlock(&fs_info
->tree_log_mutex
);
2231 mutex_unlock(&fs_info
->reloc_mutex
);
2232 goto scrub_continue
;
2235 btrfs_prepare_extent_commit(fs_info
);
2237 cur_trans
= fs_info
->running_transaction
;
2239 btrfs_set_root_node(&fs_info
->tree_root
->root_item
,
2240 fs_info
->tree_root
->node
);
2241 list_add_tail(&fs_info
->tree_root
->dirty_list
,
2242 &cur_trans
->switch_commits
);
2244 btrfs_set_root_node(&fs_info
->chunk_root
->root_item
,
2245 fs_info
->chunk_root
->node
);
2246 list_add_tail(&fs_info
->chunk_root
->dirty_list
,
2247 &cur_trans
->switch_commits
);
2249 switch_commit_roots(cur_trans
, fs_info
);
2251 ASSERT(list_empty(&cur_trans
->dirty_bgs
));
2252 ASSERT(list_empty(&cur_trans
->io_bgs
));
2253 update_super_roots(fs_info
);
2255 btrfs_set_super_log_root(fs_info
->super_copy
, 0);
2256 btrfs_set_super_log_root_level(fs_info
->super_copy
, 0);
2257 memcpy(fs_info
->super_for_commit
, fs_info
->super_copy
,
2258 sizeof(*fs_info
->super_copy
));
2260 btrfs_update_commit_device_size(fs_info
);
2261 btrfs_update_commit_device_bytes_used(fs_info
, cur_trans
);
2263 clear_bit(BTRFS_FS_LOG1_ERR
, &fs_info
->flags
);
2264 clear_bit(BTRFS_FS_LOG2_ERR
, &fs_info
->flags
);
2266 btrfs_trans_release_chunk_metadata(trans
);
2268 spin_lock(&fs_info
->trans_lock
);
2269 cur_trans
->state
= TRANS_STATE_UNBLOCKED
;
2270 fs_info
->running_transaction
= NULL
;
2271 spin_unlock(&fs_info
->trans_lock
);
2272 mutex_unlock(&fs_info
->reloc_mutex
);
2274 wake_up(&fs_info
->transaction_wait
);
2276 ret
= btrfs_write_and_wait_transaction(trans
, fs_info
);
2278 btrfs_handle_fs_error(fs_info
, ret
,
2279 "Error while writing out transaction");
2280 mutex_unlock(&fs_info
->tree_log_mutex
);
2281 goto scrub_continue
;
2284 ret
= write_all_supers(fs_info
, 0);
2286 mutex_unlock(&fs_info
->tree_log_mutex
);
2287 goto scrub_continue
;
2291 * the super is written, we can safely allow the tree-loggers
2292 * to go about their business
2294 mutex_unlock(&fs_info
->tree_log_mutex
);
2296 btrfs_finish_extent_commit(trans
, fs_info
);
2298 if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS
, &cur_trans
->flags
))
2299 btrfs_clear_space_info_full(fs_info
);
2301 fs_info
->last_trans_committed
= cur_trans
->transid
;
2303 * We needn't acquire the lock here because there is no other task
2304 * which can change it.
2306 cur_trans
->state
= TRANS_STATE_COMPLETED
;
2307 wake_up(&cur_trans
->commit_wait
);
2309 spin_lock(&fs_info
->trans_lock
);
2310 list_del_init(&cur_trans
->list
);
2311 spin_unlock(&fs_info
->trans_lock
);
2313 btrfs_put_transaction(cur_trans
);
2314 btrfs_put_transaction(cur_trans
);
2316 if (trans
->type
& __TRANS_FREEZABLE
)
2317 sb_end_intwrite(fs_info
->sb
);
2319 trace_btrfs_transaction_commit(trans
->root
);
2321 btrfs_scrub_continue(fs_info
);
2323 if (current
->journal_info
== trans
)
2324 current
->journal_info
= NULL
;
2326 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
2331 btrfs_scrub_continue(fs_info
);
2332 cleanup_transaction
:
2333 btrfs_trans_release_metadata(trans
, fs_info
);
2334 btrfs_trans_release_chunk_metadata(trans
);
2335 trans
->block_rsv
= NULL
;
2336 btrfs_warn(fs_info
, "Skipping commit of aborted transaction.");
2337 if (current
->journal_info
== trans
)
2338 current
->journal_info
= NULL
;
2339 cleanup_transaction(trans
, trans
->root
, ret
);
2345 * return < 0 if error
2346 * 0 if there are no more dead_roots at the time of call
2347 * 1 there are more to be processed, call me again
2349 * The return value indicates there are certainly more snapshots to delete, but
2350 * if there comes a new one during processing, it may return 0. We don't mind,
2351 * because btrfs_commit_super will poke cleaner thread and it will process it a
2352 * few seconds later.
2354 int btrfs_clean_one_deleted_snapshot(struct btrfs_root
*root
)
2357 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2359 spin_lock(&fs_info
->trans_lock
);
2360 if (list_empty(&fs_info
->dead_roots
)) {
2361 spin_unlock(&fs_info
->trans_lock
);
2364 root
= list_first_entry(&fs_info
->dead_roots
,
2365 struct btrfs_root
, root_list
);
2366 list_del_init(&root
->root_list
);
2367 spin_unlock(&fs_info
->trans_lock
);
2369 btrfs_debug(fs_info
, "cleaner removing %llu", root
->objectid
);
2371 btrfs_kill_all_delayed_nodes(root
);
2373 if (btrfs_header_backref_rev(root
->node
) <
2374 BTRFS_MIXED_BACKREF_REV
)
2375 ret
= btrfs_drop_snapshot(root
, NULL
, 0, 0);
2377 ret
= btrfs_drop_snapshot(root
, NULL
, 1, 0);
2379 return (ret
< 0) ? 0 : 1;
2382 void btrfs_apply_pending_changes(struct btrfs_fs_info
*fs_info
)
2387 prev
= xchg(&fs_info
->pending_changes
, 0);
2391 bit
= 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE
;
2393 btrfs_set_opt(fs_info
->mount_opt
, INODE_MAP_CACHE
);
2396 bit
= 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE
;
2398 btrfs_clear_opt(fs_info
->mount_opt
, INODE_MAP_CACHE
);
2401 bit
= 1 << BTRFS_PENDING_COMMIT
;
2403 btrfs_debug(fs_info
, "pending commit done");
2408 "unknown pending changes left 0x%lx, ignoring", prev
);