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 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
);
96 kmem_cache_free(btrfs_transaction_cachep
, transaction
);
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 atomic_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
= kmem_cache_alloc(btrfs_transaction_cachep
, 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
241 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
243 } else if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
244 spin_unlock(&fs_info
->trans_lock
);
245 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
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 atomic_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
->i_mapping
);
298 fs_info
->generation
++;
299 cur_trans
->transid
= fs_info
->generation
;
300 fs_info
->running_transaction
= cur_trans
;
301 cur_trans
->aborted
= 0;
302 spin_unlock(&fs_info
->trans_lock
);
308 * this does all the record keeping required to make sure that a reference
309 * counted root is properly recorded in a given transaction. This is required
310 * to make sure the old root from before we joined the transaction is deleted
311 * when the transaction commits
313 static int record_root_in_trans(struct btrfs_trans_handle
*trans
,
314 struct btrfs_root
*root
,
317 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(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 atomic_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
)
479 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
481 struct btrfs_trans_handle
*h
;
482 struct btrfs_transaction
*cur_trans
;
484 u64 qgroup_reserved
= 0;
485 bool reloc_reserved
= false;
488 /* Send isn't supposed to start transactions. */
489 ASSERT(current
->journal_info
!= BTRFS_SEND_TRANS_STUB
);
491 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
492 return ERR_PTR(-EROFS
);
494 if (current
->journal_info
) {
495 WARN_ON(type
& TRANS_EXTWRITERS
);
496 h
= current
->journal_info
;
498 WARN_ON(h
->use_count
> 2);
499 h
->orig_rsv
= h
->block_rsv
;
505 * Do the reservation before we join the transaction so we can do all
506 * the appropriate flushing if need be.
508 if (num_items
> 0 && root
!= fs_info
->chunk_root
) {
509 qgroup_reserved
= num_items
* fs_info
->nodesize
;
510 ret
= btrfs_qgroup_reserve_meta(root
, qgroup_reserved
);
514 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, num_items
);
516 * Do the reservation for the relocation root creation
518 if (need_reserve_reloc_root(root
)) {
519 num_bytes
+= fs_info
->nodesize
;
520 reloc_reserved
= true;
523 ret
= btrfs_block_rsv_add(root
, &fs_info
->trans_block_rsv
,
529 h
= kmem_cache_zalloc(btrfs_trans_handle_cachep
, GFP_NOFS
);
536 * If we are JOIN_NOLOCK we're already committing a transaction and
537 * waiting on this guy, so we don't need to do the sb_start_intwrite
538 * because we're already holding a ref. We need this because we could
539 * have raced in and did an fsync() on a file which can kick a commit
540 * and then we deadlock with somebody doing a freeze.
542 * If we are ATTACH, it means we just want to catch the current
543 * transaction and commit it, so we needn't do sb_start_intwrite().
545 if (type
& __TRANS_FREEZABLE
)
546 sb_start_intwrite(fs_info
->sb
);
548 if (may_wait_transaction(fs_info
, type
))
549 wait_current_trans(fs_info
);
552 ret
= join_transaction(fs_info
, type
);
554 wait_current_trans(fs_info
);
555 if (unlikely(type
== TRANS_ATTACH
))
558 } while (ret
== -EBUSY
);
563 cur_trans
= fs_info
->running_transaction
;
565 h
->transid
= cur_trans
->transid
;
566 h
->transaction
= cur_trans
;
569 h
->fs_info
= root
->fs_info
;
572 h
->can_flush_pending_bgs
= true;
573 INIT_LIST_HEAD(&h
->qgroup_ref_list
);
574 INIT_LIST_HEAD(&h
->new_bgs
);
577 if (cur_trans
->state
>= TRANS_STATE_BLOCKED
&&
578 may_wait_transaction(fs_info
, type
)) {
579 current
->journal_info
= h
;
580 btrfs_commit_transaction(h
, root
);
585 trace_btrfs_space_reservation(fs_info
, "transaction",
586 h
->transid
, num_bytes
, 1);
587 h
->block_rsv
= &fs_info
->trans_block_rsv
;
588 h
->bytes_reserved
= num_bytes
;
589 h
->reloc_reserved
= reloc_reserved
;
593 btrfs_record_root_in_trans(h
, root
);
595 if (!current
->journal_info
&& type
!= TRANS_USERSPACE
)
596 current
->journal_info
= h
;
600 if (type
& __TRANS_FREEZABLE
)
601 sb_end_intwrite(fs_info
->sb
);
602 kmem_cache_free(btrfs_trans_handle_cachep
, h
);
605 btrfs_block_rsv_release(fs_info
, &fs_info
->trans_block_rsv
,
608 btrfs_qgroup_free_meta(root
, qgroup_reserved
);
612 struct btrfs_trans_handle
*btrfs_start_transaction(struct btrfs_root
*root
,
613 unsigned int num_items
)
615 return start_transaction(root
, num_items
, TRANS_START
,
616 BTRFS_RESERVE_FLUSH_ALL
);
618 struct btrfs_trans_handle
*btrfs_start_transaction_fallback_global_rsv(
619 struct btrfs_root
*root
,
620 unsigned int num_items
,
623 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
624 struct btrfs_trans_handle
*trans
;
628 trans
= btrfs_start_transaction(root
, num_items
);
629 if (!IS_ERR(trans
) || PTR_ERR(trans
) != -ENOSPC
)
632 trans
= btrfs_start_transaction(root
, 0);
636 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, num_items
);
637 ret
= btrfs_cond_migrate_bytes(fs_info
, &fs_info
->trans_block_rsv
,
638 num_bytes
, min_factor
);
640 btrfs_end_transaction(trans
, root
);
644 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
645 trans
->bytes_reserved
= num_bytes
;
646 trace_btrfs_space_reservation(fs_info
, "transaction",
647 trans
->transid
, num_bytes
, 1);
652 struct btrfs_trans_handle
*btrfs_start_transaction_lflush(
653 struct btrfs_root
*root
,
654 unsigned int num_items
)
656 return start_transaction(root
, num_items
, TRANS_START
,
657 BTRFS_RESERVE_FLUSH_LIMIT
);
660 struct btrfs_trans_handle
*btrfs_join_transaction(struct btrfs_root
*root
)
662 return start_transaction(root
, 0, TRANS_JOIN
,
663 BTRFS_RESERVE_NO_FLUSH
);
666 struct btrfs_trans_handle
*btrfs_join_transaction_nolock(struct btrfs_root
*root
)
668 return start_transaction(root
, 0, TRANS_JOIN_NOLOCK
,
669 BTRFS_RESERVE_NO_FLUSH
);
672 struct btrfs_trans_handle
*btrfs_start_ioctl_transaction(struct btrfs_root
*root
)
674 return start_transaction(root
, 0, TRANS_USERSPACE
,
675 BTRFS_RESERVE_NO_FLUSH
);
679 * btrfs_attach_transaction() - catch the running transaction
681 * It is used when we want to commit the current the transaction, but
682 * don't want to start a new one.
684 * Note: If this function return -ENOENT, it just means there is no
685 * running transaction. But it is possible that the inactive transaction
686 * is still in the memory, not fully on disk. If you hope there is no
687 * inactive transaction in the fs when -ENOENT is returned, you should
689 * btrfs_attach_transaction_barrier()
691 struct btrfs_trans_handle
*btrfs_attach_transaction(struct btrfs_root
*root
)
693 return start_transaction(root
, 0, TRANS_ATTACH
,
694 BTRFS_RESERVE_NO_FLUSH
);
698 * btrfs_attach_transaction_barrier() - catch the running transaction
700 * It is similar to the above function, the differentia is this one
701 * will wait for all the inactive transactions until they fully
704 struct btrfs_trans_handle
*
705 btrfs_attach_transaction_barrier(struct btrfs_root
*root
)
707 struct btrfs_trans_handle
*trans
;
709 trans
= start_transaction(root
, 0, TRANS_ATTACH
,
710 BTRFS_RESERVE_NO_FLUSH
);
711 if (IS_ERR(trans
) && PTR_ERR(trans
) == -ENOENT
)
712 btrfs_wait_for_commit(root
->fs_info
, 0);
717 /* wait for a transaction commit to be fully complete */
718 static noinline
void wait_for_commit(struct btrfs_transaction
*commit
)
720 wait_event(commit
->commit_wait
, commit
->state
== TRANS_STATE_COMPLETED
);
723 int btrfs_wait_for_commit(struct btrfs_fs_info
*fs_info
, u64 transid
)
725 struct btrfs_transaction
*cur_trans
= NULL
, *t
;
729 if (transid
<= fs_info
->last_trans_committed
)
732 /* find specified transaction */
733 spin_lock(&fs_info
->trans_lock
);
734 list_for_each_entry(t
, &fs_info
->trans_list
, list
) {
735 if (t
->transid
== transid
) {
737 atomic_inc(&cur_trans
->use_count
);
741 if (t
->transid
> transid
) {
746 spin_unlock(&fs_info
->trans_lock
);
749 * The specified transaction doesn't exist, or we
750 * raced with btrfs_commit_transaction
753 if (transid
> fs_info
->last_trans_committed
)
758 /* find newest transaction that is committing | committed */
759 spin_lock(&fs_info
->trans_lock
);
760 list_for_each_entry_reverse(t
, &fs_info
->trans_list
,
762 if (t
->state
>= TRANS_STATE_COMMIT_START
) {
763 if (t
->state
== TRANS_STATE_COMPLETED
)
766 atomic_inc(&cur_trans
->use_count
);
770 spin_unlock(&fs_info
->trans_lock
);
772 goto out
; /* nothing committing|committed */
775 wait_for_commit(cur_trans
);
776 btrfs_put_transaction(cur_trans
);
781 void btrfs_throttle(struct btrfs_fs_info
*fs_info
)
783 if (!atomic_read(&fs_info
->open_ioctl_trans
))
784 wait_current_trans(fs_info
);
787 static int should_end_transaction(struct btrfs_trans_handle
*trans
)
789 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
791 if (fs_info
->global_block_rsv
.space_info
->full
&&
792 btrfs_check_space_for_delayed_refs(trans
, fs_info
))
795 return !!btrfs_block_rsv_check(&fs_info
->global_block_rsv
, 5);
798 int btrfs_should_end_transaction(struct btrfs_trans_handle
*trans
,
799 struct btrfs_root
*root
)
801 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
802 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
807 if (cur_trans
->state
>= TRANS_STATE_BLOCKED
||
808 cur_trans
->delayed_refs
.flushing
)
811 updates
= trans
->delayed_ref_updates
;
812 trans
->delayed_ref_updates
= 0;
814 err
= btrfs_run_delayed_refs(trans
, fs_info
, updates
* 2);
815 if (err
) /* Error code will also eval true */
819 return should_end_transaction(trans
);
822 static int __btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
823 struct btrfs_root
*root
, int throttle
)
825 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
826 struct btrfs_fs_info
*info
= root
->fs_info
;
827 u64 transid
= trans
->transid
;
828 unsigned long cur
= trans
->delayed_ref_updates
;
829 int lock
= (trans
->type
!= TRANS_JOIN_NOLOCK
);
831 int must_run_delayed_refs
= 0;
833 if (trans
->use_count
> 1) {
835 trans
->block_rsv
= trans
->orig_rsv
;
839 btrfs_trans_release_metadata(trans
, info
);
840 trans
->block_rsv
= NULL
;
842 if (!list_empty(&trans
->new_bgs
))
843 btrfs_create_pending_block_groups(trans
, info
);
845 trans
->delayed_ref_updates
= 0;
847 must_run_delayed_refs
=
848 btrfs_should_throttle_delayed_refs(trans
, info
);
849 cur
= max_t(unsigned long, cur
, 32);
852 * don't make the caller wait if they are from a NOLOCK
853 * or ATTACH transaction, it will deadlock with commit
855 if (must_run_delayed_refs
== 1 &&
856 (trans
->type
& (__TRANS_JOIN_NOLOCK
| __TRANS_ATTACH
)))
857 must_run_delayed_refs
= 2;
860 btrfs_trans_release_metadata(trans
, info
);
861 trans
->block_rsv
= NULL
;
863 if (!list_empty(&trans
->new_bgs
))
864 btrfs_create_pending_block_groups(trans
, info
);
866 btrfs_trans_release_chunk_metadata(trans
);
868 if (lock
&& !atomic_read(&info
->open_ioctl_trans
) &&
869 should_end_transaction(trans
) &&
870 ACCESS_ONCE(cur_trans
->state
) == TRANS_STATE_RUNNING
) {
871 spin_lock(&info
->trans_lock
);
872 if (cur_trans
->state
== TRANS_STATE_RUNNING
)
873 cur_trans
->state
= TRANS_STATE_BLOCKED
;
874 spin_unlock(&info
->trans_lock
);
877 if (lock
&& ACCESS_ONCE(cur_trans
->state
) == TRANS_STATE_BLOCKED
) {
879 return btrfs_commit_transaction(trans
, root
);
881 wake_up_process(info
->transaction_kthread
);
884 if (trans
->type
& __TRANS_FREEZABLE
)
885 sb_end_intwrite(info
->sb
);
887 WARN_ON(cur_trans
!= info
->running_transaction
);
888 WARN_ON(atomic_read(&cur_trans
->num_writers
) < 1);
889 atomic_dec(&cur_trans
->num_writers
);
890 extwriter_counter_dec(cur_trans
, trans
->type
);
893 * Make sure counter is updated before we wake up waiters.
896 if (waitqueue_active(&cur_trans
->writer_wait
))
897 wake_up(&cur_trans
->writer_wait
);
898 btrfs_put_transaction(cur_trans
);
900 if (current
->journal_info
== trans
)
901 current
->journal_info
= NULL
;
904 btrfs_run_delayed_iputs(info
);
906 if (trans
->aborted
||
907 test_bit(BTRFS_FS_STATE_ERROR
, &info
->fs_state
)) {
908 wake_up_process(info
->transaction_kthread
);
911 assert_qgroups_uptodate(trans
);
913 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
914 if (must_run_delayed_refs
) {
915 btrfs_async_run_delayed_refs(info
, cur
, transid
,
916 must_run_delayed_refs
== 1);
921 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
922 struct btrfs_root
*root
)
924 return __btrfs_end_transaction(trans
, root
, 0);
927 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
,
928 struct btrfs_root
*root
)
930 return __btrfs_end_transaction(trans
, root
, 1);
934 * when btree blocks are allocated, they have some corresponding bits set for
935 * them in one of two extent_io trees. This is used to make sure all of
936 * those extents are sent to disk but does not wait on them
938 int btrfs_write_marked_extents(struct btrfs_fs_info
*fs_info
,
939 struct extent_io_tree
*dirty_pages
, int mark
)
943 struct address_space
*mapping
= fs_info
->btree_inode
->i_mapping
;
944 struct extent_state
*cached_state
= NULL
;
948 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
949 mark
, &cached_state
)) {
950 bool wait_writeback
= false;
952 err
= convert_extent_bit(dirty_pages
, start
, end
,
954 mark
, &cached_state
);
956 * convert_extent_bit can return -ENOMEM, which is most of the
957 * time a temporary error. So when it happens, ignore the error
958 * and wait for writeback of this range to finish - because we
959 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
960 * to btrfs_wait_marked_extents() would not know that writeback
961 * for this range started and therefore wouldn't wait for it to
962 * finish - we don't want to commit a superblock that points to
963 * btree nodes/leafs for which writeback hasn't finished yet
964 * (and without errors).
965 * We cleanup any entries left in the io tree when committing
966 * the transaction (through clear_btree_io_tree()).
968 if (err
== -ENOMEM
) {
970 wait_writeback
= true;
973 err
= filemap_fdatawrite_range(mapping
, start
, end
);
976 else if (wait_writeback
)
977 werr
= filemap_fdatawait_range(mapping
, start
, end
);
978 free_extent_state(cached_state
);
987 * when btree blocks are allocated, they have some corresponding bits set for
988 * them in one of two extent_io trees. This is used to make sure all of
989 * those extents are on disk for transaction or log commit. We wait
990 * on all the pages and clear them from the dirty pages state tree
992 int btrfs_wait_marked_extents(struct btrfs_root
*root
,
993 struct extent_io_tree
*dirty_pages
, int mark
)
997 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
998 struct address_space
*mapping
= fs_info
->btree_inode
->i_mapping
;
999 struct extent_state
*cached_state
= NULL
;
1002 bool errors
= false;
1004 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
1005 EXTENT_NEED_WAIT
, &cached_state
)) {
1007 * Ignore -ENOMEM errors returned by clear_extent_bit().
1008 * When committing the transaction, we'll remove any entries
1009 * left in the io tree. For a log commit, we don't remove them
1010 * after committing the log because the tree can be accessed
1011 * concurrently - we do it only at transaction commit time when
1012 * it's safe to do it (through clear_btree_io_tree()).
1014 err
= clear_extent_bit(dirty_pages
, start
, end
,
1016 0, 0, &cached_state
, GFP_NOFS
);
1020 err
= filemap_fdatawait_range(mapping
, start
, end
);
1023 free_extent_state(cached_state
);
1024 cached_state
= NULL
;
1031 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
1032 if ((mark
& EXTENT_DIRTY
) &&
1033 test_and_clear_bit(BTRFS_FS_LOG1_ERR
, &fs_info
->flags
))
1036 if ((mark
& EXTENT_NEW
) &&
1037 test_and_clear_bit(BTRFS_FS_LOG2_ERR
, &fs_info
->flags
))
1040 if (test_and_clear_bit(BTRFS_FS_BTREE_ERR
, &fs_info
->flags
))
1044 if (errors
&& !werr
)
1051 * when btree blocks are allocated, they have some corresponding bits set for
1052 * them in one of two extent_io trees. This is used to make sure all of
1053 * those extents are on disk for transaction or log commit
1055 static int btrfs_write_and_wait_marked_extents(struct btrfs_root
*root
,
1056 struct extent_io_tree
*dirty_pages
, int mark
)
1060 struct blk_plug plug
;
1062 blk_start_plug(&plug
);
1063 ret
= btrfs_write_marked_extents(root
->fs_info
, dirty_pages
, mark
);
1064 blk_finish_plug(&plug
);
1065 ret2
= btrfs_wait_marked_extents(root
, dirty_pages
, mark
);
1074 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
,
1075 struct btrfs_root
*root
)
1079 ret
= btrfs_write_and_wait_marked_extents(root
,
1080 &trans
->transaction
->dirty_pages
,
1082 clear_btree_io_tree(&trans
->transaction
->dirty_pages
);
1088 * this is used to update the root pointer in the tree of tree roots.
1090 * But, in the case of the extent allocation tree, updating the root
1091 * pointer may allocate blocks which may change the root of the extent
1094 * So, this loops and repeats and makes sure the cowonly root didn't
1095 * change while the root pointer was being updated in the metadata.
1097 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
1098 struct btrfs_root
*root
)
1101 u64 old_root_bytenr
;
1103 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1104 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1106 old_root_used
= btrfs_root_used(&root
->root_item
);
1109 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
1110 if (old_root_bytenr
== root
->node
->start
&&
1111 old_root_used
== btrfs_root_used(&root
->root_item
))
1114 btrfs_set_root_node(&root
->root_item
, root
->node
);
1115 ret
= btrfs_update_root(trans
, tree_root
,
1121 old_root_used
= btrfs_root_used(&root
->root_item
);
1128 * update all the cowonly tree roots on disk
1130 * The error handling in this function may not be obvious. Any of the
1131 * failures will cause the file system to go offline. We still need
1132 * to clean up the delayed refs.
1134 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
,
1135 struct btrfs_fs_info
*fs_info
)
1137 struct list_head
*dirty_bgs
= &trans
->transaction
->dirty_bgs
;
1138 struct list_head
*io_bgs
= &trans
->transaction
->io_bgs
;
1139 struct list_head
*next
;
1140 struct extent_buffer
*eb
;
1143 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
1144 ret
= btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
,
1146 btrfs_tree_unlock(eb
);
1147 free_extent_buffer(eb
);
1152 ret
= btrfs_run_delayed_refs(trans
, fs_info
, (unsigned long)-1);
1156 ret
= btrfs_run_dev_stats(trans
, fs_info
);
1159 ret
= btrfs_run_dev_replace(trans
, fs_info
);
1162 ret
= btrfs_run_qgroups(trans
, fs_info
);
1166 ret
= btrfs_setup_space_cache(trans
, fs_info
);
1170 /* run_qgroups might have added some more refs */
1171 ret
= btrfs_run_delayed_refs(trans
, fs_info
, (unsigned long)-1);
1175 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
1176 struct btrfs_root
*root
;
1177 next
= fs_info
->dirty_cowonly_roots
.next
;
1178 list_del_init(next
);
1179 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
1180 clear_bit(BTRFS_ROOT_DIRTY
, &root
->state
);
1182 if (root
!= fs_info
->extent_root
)
1183 list_add_tail(&root
->dirty_list
,
1184 &trans
->transaction
->switch_commits
);
1185 ret
= update_cowonly_root(trans
, root
);
1188 ret
= btrfs_run_delayed_refs(trans
, fs_info
, (unsigned long)-1);
1193 while (!list_empty(dirty_bgs
) || !list_empty(io_bgs
)) {
1194 ret
= btrfs_write_dirty_block_groups(trans
, fs_info
);
1197 ret
= btrfs_run_delayed_refs(trans
, fs_info
, (unsigned long)-1);
1202 if (!list_empty(&fs_info
->dirty_cowonly_roots
))
1205 list_add_tail(&fs_info
->extent_root
->dirty_list
,
1206 &trans
->transaction
->switch_commits
);
1207 btrfs_after_dev_replace_commit(fs_info
);
1213 * dead roots are old snapshots that need to be deleted. This allocates
1214 * a dirty root struct and adds it into the list of dead roots that need to
1217 void btrfs_add_dead_root(struct btrfs_root
*root
)
1219 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1221 spin_lock(&fs_info
->trans_lock
);
1222 if (list_empty(&root
->root_list
))
1223 list_add_tail(&root
->root_list
, &fs_info
->dead_roots
);
1224 spin_unlock(&fs_info
->trans_lock
);
1228 * update all the cowonly tree roots on disk
1230 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
,
1231 struct btrfs_fs_info
*fs_info
)
1233 struct btrfs_root
*gang
[8];
1238 spin_lock(&fs_info
->fs_roots_radix_lock
);
1240 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
1243 BTRFS_ROOT_TRANS_TAG
);
1246 for (i
= 0; i
< ret
; i
++) {
1247 struct btrfs_root
*root
= gang
[i
];
1248 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
1249 (unsigned long)root
->root_key
.objectid
,
1250 BTRFS_ROOT_TRANS_TAG
);
1251 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1253 btrfs_free_log(trans
, root
);
1254 btrfs_update_reloc_root(trans
, root
);
1255 btrfs_orphan_commit_root(trans
, root
);
1257 btrfs_save_ino_cache(root
, trans
);
1259 /* see comments in should_cow_block() */
1260 clear_bit(BTRFS_ROOT_FORCE_COW
, &root
->state
);
1261 smp_mb__after_atomic();
1263 if (root
->commit_root
!= root
->node
) {
1264 list_add_tail(&root
->dirty_list
,
1265 &trans
->transaction
->switch_commits
);
1266 btrfs_set_root_node(&root
->root_item
,
1270 err
= btrfs_update_root(trans
, fs_info
->tree_root
,
1273 spin_lock(&fs_info
->fs_roots_radix_lock
);
1276 btrfs_qgroup_free_meta_all(root
);
1279 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1284 * defrag a given btree.
1285 * Every leaf in the btree is read and defragged.
1287 int btrfs_defrag_root(struct btrfs_root
*root
)
1289 struct btrfs_fs_info
*info
= root
->fs_info
;
1290 struct btrfs_trans_handle
*trans
;
1293 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING
, &root
->state
))
1297 trans
= btrfs_start_transaction(root
, 0);
1299 return PTR_ERR(trans
);
1301 ret
= btrfs_defrag_leaves(trans
, root
);
1303 btrfs_end_transaction(trans
, root
);
1304 btrfs_btree_balance_dirty(info
);
1307 if (btrfs_fs_closing(info
) || ret
!= -EAGAIN
)
1310 if (btrfs_defrag_cancelled(info
)) {
1311 btrfs_debug(info
, "defrag_root cancelled");
1316 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING
, &root
->state
);
1321 * Do all special snapshot related qgroup dirty hack.
1323 * Will do all needed qgroup inherit and dirty hack like switch commit
1324 * roots inside one transaction and write all btree into disk, to make
1327 static int qgroup_account_snapshot(struct btrfs_trans_handle
*trans
,
1328 struct btrfs_root
*src
,
1329 struct btrfs_root
*parent
,
1330 struct btrfs_qgroup_inherit
*inherit
,
1333 struct btrfs_fs_info
*fs_info
= src
->fs_info
;
1337 * Save some performance in the case that qgroups are not
1338 * enabled. If this check races with the ioctl, rescan will
1341 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
1342 if (!test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
)) {
1343 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1346 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1349 * We are going to commit transaction, see btrfs_commit_transaction()
1350 * comment for reason locking tree_log_mutex
1352 mutex_lock(&fs_info
->tree_log_mutex
);
1354 ret
= commit_fs_roots(trans
, fs_info
);
1357 ret
= btrfs_qgroup_prepare_account_extents(trans
, fs_info
);
1360 ret
= btrfs_qgroup_account_extents(trans
, fs_info
);
1364 /* Now qgroup are all updated, we can inherit it to new qgroups */
1365 ret
= btrfs_qgroup_inherit(trans
, fs_info
,
1366 src
->root_key
.objectid
, dst_objectid
,
1372 * Now we do a simplified commit transaction, which will:
1373 * 1) commit all subvolume and extent tree
1374 * To ensure all subvolume and extent tree have a valid
1375 * commit_root to accounting later insert_dir_item()
1376 * 2) write all btree blocks onto disk
1377 * This is to make sure later btree modification will be cowed
1378 * Or commit_root can be populated and cause wrong qgroup numbers
1379 * In this simplified commit, we don't really care about other trees
1380 * like chunk and root tree, as they won't affect qgroup.
1381 * And we don't write super to avoid half committed status.
1383 ret
= commit_cowonly_roots(trans
, fs_info
);
1386 switch_commit_roots(trans
->transaction
, fs_info
);
1387 ret
= btrfs_write_and_wait_transaction(trans
, src
);
1389 btrfs_handle_fs_error(fs_info
, ret
,
1390 "Error while writing out transaction for qgroup");
1393 mutex_unlock(&fs_info
->tree_log_mutex
);
1396 * Force parent root to be updated, as we recorded it before so its
1397 * last_trans == cur_transid.
1398 * Or it won't be committed again onto disk after later
1402 record_root_in_trans(trans
, parent
, 1);
1407 * new snapshots need to be created at a very specific time in the
1408 * transaction commit. This does the actual creation.
1411 * If the error which may affect the commitment of the current transaction
1412 * happens, we should return the error number. If the error which just affect
1413 * the creation of the pending snapshots, just return 0.
1415 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
1416 struct btrfs_fs_info
*fs_info
,
1417 struct btrfs_pending_snapshot
*pending
)
1419 struct btrfs_key key
;
1420 struct btrfs_root_item
*new_root_item
;
1421 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1422 struct btrfs_root
*root
= pending
->root
;
1423 struct btrfs_root
*parent_root
;
1424 struct btrfs_block_rsv
*rsv
;
1425 struct inode
*parent_inode
;
1426 struct btrfs_path
*path
;
1427 struct btrfs_dir_item
*dir_item
;
1428 struct dentry
*dentry
;
1429 struct extent_buffer
*tmp
;
1430 struct extent_buffer
*old
;
1431 struct timespec cur_time
;
1439 ASSERT(pending
->path
);
1440 path
= pending
->path
;
1442 ASSERT(pending
->root_item
);
1443 new_root_item
= pending
->root_item
;
1445 pending
->error
= btrfs_find_free_objectid(tree_root
, &objectid
);
1447 goto no_free_objectid
;
1450 * Make qgroup to skip current new snapshot's qgroupid, as it is
1451 * accounted by later btrfs_qgroup_inherit().
1453 btrfs_set_skip_qgroup(trans
, objectid
);
1455 btrfs_reloc_pre_snapshot(pending
, &to_reserve
);
1457 if (to_reserve
> 0) {
1458 pending
->error
= btrfs_block_rsv_add(root
,
1459 &pending
->block_rsv
,
1461 BTRFS_RESERVE_NO_FLUSH
);
1463 goto clear_skip_qgroup
;
1466 key
.objectid
= objectid
;
1467 key
.offset
= (u64
)-1;
1468 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1470 rsv
= trans
->block_rsv
;
1471 trans
->block_rsv
= &pending
->block_rsv
;
1472 trans
->bytes_reserved
= trans
->block_rsv
->reserved
;
1473 trace_btrfs_space_reservation(fs_info
, "transaction",
1475 trans
->bytes_reserved
, 1);
1476 dentry
= pending
->dentry
;
1477 parent_inode
= pending
->dir
;
1478 parent_root
= BTRFS_I(parent_inode
)->root
;
1479 record_root_in_trans(trans
, parent_root
, 0);
1481 cur_time
= current_time(parent_inode
);
1484 * insert the directory item
1486 ret
= btrfs_set_inode_index(parent_inode
, &index
);
1487 BUG_ON(ret
); /* -ENOMEM */
1489 /* check if there is a file/dir which has the same name. */
1490 dir_item
= btrfs_lookup_dir_item(NULL
, parent_root
, path
,
1491 btrfs_ino(parent_inode
),
1492 dentry
->d_name
.name
,
1493 dentry
->d_name
.len
, 0);
1494 if (dir_item
!= NULL
&& !IS_ERR(dir_item
)) {
1495 pending
->error
= -EEXIST
;
1496 goto dir_item_existed
;
1497 } else if (IS_ERR(dir_item
)) {
1498 ret
= PTR_ERR(dir_item
);
1499 btrfs_abort_transaction(trans
, ret
);
1502 btrfs_release_path(path
);
1505 * pull in the delayed directory update
1506 * and the delayed inode item
1507 * otherwise we corrupt the FS during
1510 ret
= btrfs_run_delayed_items(trans
, fs_info
);
1511 if (ret
) { /* Transaction aborted */
1512 btrfs_abort_transaction(trans
, ret
);
1516 record_root_in_trans(trans
, root
, 0);
1517 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
1518 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
1519 btrfs_check_and_init_root_item(new_root_item
);
1521 root_flags
= btrfs_root_flags(new_root_item
);
1522 if (pending
->readonly
)
1523 root_flags
|= BTRFS_ROOT_SUBVOL_RDONLY
;
1525 root_flags
&= ~BTRFS_ROOT_SUBVOL_RDONLY
;
1526 btrfs_set_root_flags(new_root_item
, root_flags
);
1528 btrfs_set_root_generation_v2(new_root_item
,
1530 uuid_le_gen(&new_uuid
);
1531 memcpy(new_root_item
->uuid
, new_uuid
.b
, BTRFS_UUID_SIZE
);
1532 memcpy(new_root_item
->parent_uuid
, root
->root_item
.uuid
,
1534 if (!(root_flags
& BTRFS_ROOT_SUBVOL_RDONLY
)) {
1535 memset(new_root_item
->received_uuid
, 0,
1536 sizeof(new_root_item
->received_uuid
));
1537 memset(&new_root_item
->stime
, 0, sizeof(new_root_item
->stime
));
1538 memset(&new_root_item
->rtime
, 0, sizeof(new_root_item
->rtime
));
1539 btrfs_set_root_stransid(new_root_item
, 0);
1540 btrfs_set_root_rtransid(new_root_item
, 0);
1542 btrfs_set_stack_timespec_sec(&new_root_item
->otime
, cur_time
.tv_sec
);
1543 btrfs_set_stack_timespec_nsec(&new_root_item
->otime
, cur_time
.tv_nsec
);
1544 btrfs_set_root_otransid(new_root_item
, trans
->transid
);
1546 old
= btrfs_lock_root_node(root
);
1547 ret
= btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
);
1549 btrfs_tree_unlock(old
);
1550 free_extent_buffer(old
);
1551 btrfs_abort_transaction(trans
, ret
);
1555 btrfs_set_lock_blocking(old
);
1557 ret
= btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
1558 /* clean up in any case */
1559 btrfs_tree_unlock(old
);
1560 free_extent_buffer(old
);
1562 btrfs_abort_transaction(trans
, ret
);
1565 /* see comments in should_cow_block() */
1566 set_bit(BTRFS_ROOT_FORCE_COW
, &root
->state
);
1569 btrfs_set_root_node(new_root_item
, tmp
);
1570 /* record when the snapshot was created in key.offset */
1571 key
.offset
= trans
->transid
;
1572 ret
= btrfs_insert_root(trans
, tree_root
, &key
, new_root_item
);
1573 btrfs_tree_unlock(tmp
);
1574 free_extent_buffer(tmp
);
1576 btrfs_abort_transaction(trans
, ret
);
1581 * insert root back/forward references
1583 ret
= btrfs_add_root_ref(trans
, fs_info
, objectid
,
1584 parent_root
->root_key
.objectid
,
1585 btrfs_ino(parent_inode
), index
,
1586 dentry
->d_name
.name
, dentry
->d_name
.len
);
1588 btrfs_abort_transaction(trans
, ret
);
1592 key
.offset
= (u64
)-1;
1593 pending
->snap
= btrfs_read_fs_root_no_name(fs_info
, &key
);
1594 if (IS_ERR(pending
->snap
)) {
1595 ret
= PTR_ERR(pending
->snap
);
1596 btrfs_abort_transaction(trans
, ret
);
1600 ret
= btrfs_reloc_post_snapshot(trans
, pending
);
1602 btrfs_abort_transaction(trans
, ret
);
1606 ret
= btrfs_run_delayed_refs(trans
, fs_info
, (unsigned long)-1);
1608 btrfs_abort_transaction(trans
, ret
);
1613 * Do special qgroup accounting for snapshot, as we do some qgroup
1614 * snapshot hack to do fast snapshot.
1615 * To co-operate with that hack, we do hack again.
1616 * Or snapshot will be greatly slowed down by a subtree qgroup rescan
1618 ret
= qgroup_account_snapshot(trans
, root
, parent_root
,
1619 pending
->inherit
, objectid
);
1623 ret
= btrfs_insert_dir_item(trans
, parent_root
,
1624 dentry
->d_name
.name
, dentry
->d_name
.len
,
1626 BTRFS_FT_DIR
, index
);
1627 /* We have check then name at the beginning, so it is impossible. */
1628 BUG_ON(ret
== -EEXIST
|| ret
== -EOVERFLOW
);
1630 btrfs_abort_transaction(trans
, ret
);
1634 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
1635 dentry
->d_name
.len
* 2);
1636 parent_inode
->i_mtime
= parent_inode
->i_ctime
=
1637 current_time(parent_inode
);
1638 ret
= btrfs_update_inode_fallback(trans
, parent_root
, parent_inode
);
1640 btrfs_abort_transaction(trans
, ret
);
1643 ret
= btrfs_uuid_tree_add(trans
, fs_info
, new_uuid
.b
,
1644 BTRFS_UUID_KEY_SUBVOL
, objectid
);
1646 btrfs_abort_transaction(trans
, ret
);
1649 if (!btrfs_is_empty_uuid(new_root_item
->received_uuid
)) {
1650 ret
= btrfs_uuid_tree_add(trans
, fs_info
,
1651 new_root_item
->received_uuid
,
1652 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
1654 if (ret
&& ret
!= -EEXIST
) {
1655 btrfs_abort_transaction(trans
, ret
);
1660 ret
= btrfs_run_delayed_refs(trans
, fs_info
, (unsigned long)-1);
1662 btrfs_abort_transaction(trans
, ret
);
1667 pending
->error
= ret
;
1669 trans
->block_rsv
= rsv
;
1670 trans
->bytes_reserved
= 0;
1672 btrfs_clear_skip_qgroup(trans
);
1674 kfree(new_root_item
);
1675 pending
->root_item
= NULL
;
1676 btrfs_free_path(path
);
1677 pending
->path
= NULL
;
1683 * create all the snapshots we've scheduled for creation
1685 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
,
1686 struct btrfs_fs_info
*fs_info
)
1688 struct btrfs_pending_snapshot
*pending
, *next
;
1689 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
1692 list_for_each_entry_safe(pending
, next
, head
, list
) {
1693 list_del(&pending
->list
);
1694 ret
= create_pending_snapshot(trans
, fs_info
, pending
);
1701 static void update_super_roots(struct btrfs_fs_info
*fs_info
)
1703 struct btrfs_root_item
*root_item
;
1704 struct btrfs_super_block
*super
;
1706 super
= fs_info
->super_copy
;
1708 root_item
= &fs_info
->chunk_root
->root_item
;
1709 super
->chunk_root
= root_item
->bytenr
;
1710 super
->chunk_root_generation
= root_item
->generation
;
1711 super
->chunk_root_level
= root_item
->level
;
1713 root_item
= &fs_info
->tree_root
->root_item
;
1714 super
->root
= root_item
->bytenr
;
1715 super
->generation
= root_item
->generation
;
1716 super
->root_level
= root_item
->level
;
1717 if (btrfs_test_opt(fs_info
, SPACE_CACHE
))
1718 super
->cache_generation
= root_item
->generation
;
1719 if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN
, &fs_info
->flags
))
1720 super
->uuid_tree_generation
= root_item
->generation
;
1723 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
1725 struct btrfs_transaction
*trans
;
1728 spin_lock(&info
->trans_lock
);
1729 trans
= info
->running_transaction
;
1731 ret
= (trans
->state
>= TRANS_STATE_COMMIT_START
);
1732 spin_unlock(&info
->trans_lock
);
1736 int btrfs_transaction_blocked(struct btrfs_fs_info
*info
)
1738 struct btrfs_transaction
*trans
;
1741 spin_lock(&info
->trans_lock
);
1742 trans
= info
->running_transaction
;
1744 ret
= is_transaction_blocked(trans
);
1745 spin_unlock(&info
->trans_lock
);
1750 * wait for the current transaction commit to start and block subsequent
1753 static void wait_current_trans_commit_start(struct btrfs_fs_info
*fs_info
,
1754 struct btrfs_transaction
*trans
)
1756 wait_event(fs_info
->transaction_blocked_wait
,
1757 trans
->state
>= TRANS_STATE_COMMIT_START
|| trans
->aborted
);
1761 * wait for the current transaction to start and then become unblocked.
1764 static void wait_current_trans_commit_start_and_unblock(
1765 struct btrfs_fs_info
*fs_info
,
1766 struct btrfs_transaction
*trans
)
1768 wait_event(fs_info
->transaction_wait
,
1769 trans
->state
>= TRANS_STATE_UNBLOCKED
|| trans
->aborted
);
1773 * commit transactions asynchronously. once btrfs_commit_transaction_async
1774 * returns, any subsequent transaction will not be allowed to join.
1776 struct btrfs_async_commit
{
1777 struct btrfs_trans_handle
*newtrans
;
1778 struct btrfs_root
*root
;
1779 struct work_struct work
;
1782 static void do_async_commit(struct work_struct
*work
)
1784 struct btrfs_async_commit
*ac
=
1785 container_of(work
, struct btrfs_async_commit
, work
);
1788 * We've got freeze protection passed with the transaction.
1789 * Tell lockdep about it.
1791 if (ac
->newtrans
->type
& __TRANS_FREEZABLE
)
1792 __sb_writers_acquired(ac
->root
->fs_info
->sb
, SB_FREEZE_FS
);
1794 current
->journal_info
= ac
->newtrans
;
1796 btrfs_commit_transaction(ac
->newtrans
, ac
->root
);
1800 int btrfs_commit_transaction_async(struct btrfs_trans_handle
*trans
,
1801 struct btrfs_root
*root
,
1802 int wait_for_unblock
)
1804 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1805 struct btrfs_async_commit
*ac
;
1806 struct btrfs_transaction
*cur_trans
;
1808 ac
= kmalloc(sizeof(*ac
), GFP_NOFS
);
1812 INIT_WORK(&ac
->work
, do_async_commit
);
1814 ac
->newtrans
= btrfs_join_transaction(root
);
1815 if (IS_ERR(ac
->newtrans
)) {
1816 int err
= PTR_ERR(ac
->newtrans
);
1821 /* take transaction reference */
1822 cur_trans
= trans
->transaction
;
1823 atomic_inc(&cur_trans
->use_count
);
1825 btrfs_end_transaction(trans
, root
);
1828 * Tell lockdep we've released the freeze rwsem, since the
1829 * async commit thread will be the one to unlock it.
1831 if (ac
->newtrans
->type
& __TRANS_FREEZABLE
)
1832 __sb_writers_release(fs_info
->sb
, SB_FREEZE_FS
);
1834 schedule_work(&ac
->work
);
1836 /* wait for transaction to start and unblock */
1837 if (wait_for_unblock
)
1838 wait_current_trans_commit_start_and_unblock(fs_info
, cur_trans
);
1840 wait_current_trans_commit_start(fs_info
, cur_trans
);
1842 if (current
->journal_info
== trans
)
1843 current
->journal_info
= NULL
;
1845 btrfs_put_transaction(cur_trans
);
1850 static void cleanup_transaction(struct btrfs_trans_handle
*trans
,
1851 struct btrfs_root
*root
, int err
)
1853 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1854 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1857 WARN_ON(trans
->use_count
> 1);
1859 btrfs_abort_transaction(trans
, err
);
1861 spin_lock(&fs_info
->trans_lock
);
1864 * If the transaction is removed from the list, it means this
1865 * transaction has been committed successfully, so it is impossible
1866 * to call the cleanup function.
1868 BUG_ON(list_empty(&cur_trans
->list
));
1870 list_del_init(&cur_trans
->list
);
1871 if (cur_trans
== fs_info
->running_transaction
) {
1872 cur_trans
->state
= TRANS_STATE_COMMIT_DOING
;
1873 spin_unlock(&fs_info
->trans_lock
);
1874 wait_event(cur_trans
->writer_wait
,
1875 atomic_read(&cur_trans
->num_writers
) == 1);
1877 spin_lock(&fs_info
->trans_lock
);
1879 spin_unlock(&fs_info
->trans_lock
);
1881 btrfs_cleanup_one_transaction(trans
->transaction
, fs_info
);
1883 spin_lock(&fs_info
->trans_lock
);
1884 if (cur_trans
== fs_info
->running_transaction
)
1885 fs_info
->running_transaction
= NULL
;
1886 spin_unlock(&fs_info
->trans_lock
);
1888 if (trans
->type
& __TRANS_FREEZABLE
)
1889 sb_end_intwrite(fs_info
->sb
);
1890 btrfs_put_transaction(cur_trans
);
1891 btrfs_put_transaction(cur_trans
);
1893 trace_btrfs_transaction_commit(root
);
1895 if (current
->journal_info
== trans
)
1896 current
->journal_info
= NULL
;
1897 btrfs_scrub_cancel(fs_info
);
1899 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1902 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info
*fs_info
)
1904 if (btrfs_test_opt(fs_info
, FLUSHONCOMMIT
))
1905 return btrfs_start_delalloc_roots(fs_info
, 1, -1);
1909 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info
*fs_info
)
1911 if (btrfs_test_opt(fs_info
, FLUSHONCOMMIT
))
1912 btrfs_wait_ordered_roots(fs_info
, -1, 0, (u64
)-1);
1916 btrfs_wait_pending_ordered(struct btrfs_transaction
*cur_trans
)
1918 wait_event(cur_trans
->pending_wait
,
1919 atomic_read(&cur_trans
->pending_ordered
) == 0);
1922 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
,
1923 struct btrfs_root
*root
)
1925 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1926 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1927 struct btrfs_transaction
*prev_trans
= NULL
;
1930 /* Stop the commit early if ->aborted is set */
1931 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1932 ret
= cur_trans
->aborted
;
1933 btrfs_end_transaction(trans
, root
);
1937 /* make a pass through all the delayed refs we have so far
1938 * any runnings procs may add more while we are here
1940 ret
= btrfs_run_delayed_refs(trans
, fs_info
, 0);
1942 btrfs_end_transaction(trans
, root
);
1946 btrfs_trans_release_metadata(trans
, fs_info
);
1947 trans
->block_rsv
= NULL
;
1949 cur_trans
= trans
->transaction
;
1952 * set the flushing flag so procs in this transaction have to
1953 * start sending their work down.
1955 cur_trans
->delayed_refs
.flushing
= 1;
1958 if (!list_empty(&trans
->new_bgs
))
1959 btrfs_create_pending_block_groups(trans
, fs_info
);
1961 ret
= btrfs_run_delayed_refs(trans
, fs_info
, 0);
1963 btrfs_end_transaction(trans
, root
);
1967 if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN
, &cur_trans
->flags
)) {
1970 /* this mutex is also taken before trying to set
1971 * block groups readonly. We need to make sure
1972 * that nobody has set a block group readonly
1973 * after a extents from that block group have been
1974 * allocated for cache files. btrfs_set_block_group_ro
1975 * will wait for the transaction to commit if it
1976 * finds BTRFS_TRANS_DIRTY_BG_RUN set.
1978 * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure
1979 * only one process starts all the block group IO. It wouldn't
1980 * hurt to have more than one go through, but there's no
1981 * real advantage to it either.
1983 mutex_lock(&fs_info
->ro_block_group_mutex
);
1984 if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN
,
1987 mutex_unlock(&fs_info
->ro_block_group_mutex
);
1990 ret
= btrfs_start_dirty_block_groups(trans
, fs_info
);
1993 btrfs_end_transaction(trans
, root
);
1997 spin_lock(&fs_info
->trans_lock
);
1998 if (cur_trans
->state
>= TRANS_STATE_COMMIT_START
) {
1999 spin_unlock(&fs_info
->trans_lock
);
2000 atomic_inc(&cur_trans
->use_count
);
2001 ret
= btrfs_end_transaction(trans
, root
);
2003 wait_for_commit(cur_trans
);
2005 if (unlikely(cur_trans
->aborted
))
2006 ret
= cur_trans
->aborted
;
2008 btrfs_put_transaction(cur_trans
);
2013 cur_trans
->state
= TRANS_STATE_COMMIT_START
;
2014 wake_up(&fs_info
->transaction_blocked_wait
);
2016 if (cur_trans
->list
.prev
!= &fs_info
->trans_list
) {
2017 prev_trans
= list_entry(cur_trans
->list
.prev
,
2018 struct btrfs_transaction
, list
);
2019 if (prev_trans
->state
!= TRANS_STATE_COMPLETED
) {
2020 atomic_inc(&prev_trans
->use_count
);
2021 spin_unlock(&fs_info
->trans_lock
);
2023 wait_for_commit(prev_trans
);
2024 ret
= prev_trans
->aborted
;
2026 btrfs_put_transaction(prev_trans
);
2028 goto cleanup_transaction
;
2030 spin_unlock(&fs_info
->trans_lock
);
2033 spin_unlock(&fs_info
->trans_lock
);
2036 extwriter_counter_dec(cur_trans
, trans
->type
);
2038 ret
= btrfs_start_delalloc_flush(fs_info
);
2040 goto cleanup_transaction
;
2042 ret
= btrfs_run_delayed_items(trans
, fs_info
);
2044 goto cleanup_transaction
;
2046 wait_event(cur_trans
->writer_wait
,
2047 extwriter_counter_read(cur_trans
) == 0);
2049 /* some pending stuffs might be added after the previous flush. */
2050 ret
= btrfs_run_delayed_items(trans
, fs_info
);
2052 goto cleanup_transaction
;
2054 btrfs_wait_delalloc_flush(fs_info
);
2056 btrfs_wait_pending_ordered(cur_trans
);
2058 btrfs_scrub_pause(fs_info
);
2060 * Ok now we need to make sure to block out any other joins while we
2061 * commit the transaction. We could have started a join before setting
2062 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
2064 spin_lock(&fs_info
->trans_lock
);
2065 cur_trans
->state
= TRANS_STATE_COMMIT_DOING
;
2066 spin_unlock(&fs_info
->trans_lock
);
2067 wait_event(cur_trans
->writer_wait
,
2068 atomic_read(&cur_trans
->num_writers
) == 1);
2070 /* ->aborted might be set after the previous check, so check it */
2071 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
2072 ret
= cur_trans
->aborted
;
2073 goto scrub_continue
;
2076 * the reloc mutex makes sure that we stop
2077 * the balancing code from coming in and moving
2078 * extents around in the middle of the commit
2080 mutex_lock(&fs_info
->reloc_mutex
);
2083 * We needn't worry about the delayed items because we will
2084 * deal with them in create_pending_snapshot(), which is the
2085 * core function of the snapshot creation.
2087 ret
= create_pending_snapshots(trans
, fs_info
);
2089 mutex_unlock(&fs_info
->reloc_mutex
);
2090 goto scrub_continue
;
2094 * We insert the dir indexes of the snapshots and update the inode
2095 * of the snapshots' parents after the snapshot creation, so there
2096 * are some delayed items which are not dealt with. Now deal with
2099 * We needn't worry that this operation will corrupt the snapshots,
2100 * because all the tree which are snapshoted will be forced to COW
2101 * the nodes and leaves.
2103 ret
= btrfs_run_delayed_items(trans
, fs_info
);
2105 mutex_unlock(&fs_info
->reloc_mutex
);
2106 goto scrub_continue
;
2109 ret
= btrfs_run_delayed_refs(trans
, fs_info
, (unsigned long)-1);
2111 mutex_unlock(&fs_info
->reloc_mutex
);
2112 goto scrub_continue
;
2115 /* Reocrd old roots for later qgroup accounting */
2116 ret
= btrfs_qgroup_prepare_account_extents(trans
, fs_info
);
2118 mutex_unlock(&fs_info
->reloc_mutex
);
2119 goto scrub_continue
;
2123 * make sure none of the code above managed to slip in a
2126 btrfs_assert_delayed_root_empty(fs_info
);
2128 WARN_ON(cur_trans
!= trans
->transaction
);
2130 /* btrfs_commit_tree_roots is responsible for getting the
2131 * various roots consistent with each other. Every pointer
2132 * in the tree of tree roots has to point to the most up to date
2133 * root for every subvolume and other tree. So, we have to keep
2134 * the tree logging code from jumping in and changing any
2137 * At this point in the commit, there can't be any tree-log
2138 * writers, but a little lower down we drop the trans mutex
2139 * and let new people in. By holding the tree_log_mutex
2140 * from now until after the super is written, we avoid races
2141 * with the tree-log code.
2143 mutex_lock(&fs_info
->tree_log_mutex
);
2145 ret
= commit_fs_roots(trans
, fs_info
);
2147 mutex_unlock(&fs_info
->tree_log_mutex
);
2148 mutex_unlock(&fs_info
->reloc_mutex
);
2149 goto scrub_continue
;
2153 * Since the transaction is done, we can apply the pending changes
2154 * before the next transaction.
2156 btrfs_apply_pending_changes(fs_info
);
2158 /* commit_fs_roots gets rid of all the tree log roots, it is now
2159 * safe to free the root of tree log roots
2161 btrfs_free_log_root_tree(trans
, fs_info
);
2164 * Since fs roots are all committed, we can get a quite accurate
2165 * new_roots. So let's do quota accounting.
2167 ret
= btrfs_qgroup_account_extents(trans
, fs_info
);
2169 mutex_unlock(&fs_info
->tree_log_mutex
);
2170 mutex_unlock(&fs_info
->reloc_mutex
);
2171 goto scrub_continue
;
2174 ret
= commit_cowonly_roots(trans
, fs_info
);
2176 mutex_unlock(&fs_info
->tree_log_mutex
);
2177 mutex_unlock(&fs_info
->reloc_mutex
);
2178 goto scrub_continue
;
2182 * The tasks which save the space cache and inode cache may also
2183 * update ->aborted, check it.
2185 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
2186 ret
= cur_trans
->aborted
;
2187 mutex_unlock(&fs_info
->tree_log_mutex
);
2188 mutex_unlock(&fs_info
->reloc_mutex
);
2189 goto scrub_continue
;
2192 btrfs_prepare_extent_commit(trans
, fs_info
);
2194 cur_trans
= fs_info
->running_transaction
;
2196 btrfs_set_root_node(&fs_info
->tree_root
->root_item
,
2197 fs_info
->tree_root
->node
);
2198 list_add_tail(&fs_info
->tree_root
->dirty_list
,
2199 &cur_trans
->switch_commits
);
2201 btrfs_set_root_node(&fs_info
->chunk_root
->root_item
,
2202 fs_info
->chunk_root
->node
);
2203 list_add_tail(&fs_info
->chunk_root
->dirty_list
,
2204 &cur_trans
->switch_commits
);
2206 switch_commit_roots(cur_trans
, fs_info
);
2208 assert_qgroups_uptodate(trans
);
2209 ASSERT(list_empty(&cur_trans
->dirty_bgs
));
2210 ASSERT(list_empty(&cur_trans
->io_bgs
));
2211 update_super_roots(fs_info
);
2213 btrfs_set_super_log_root(fs_info
->super_copy
, 0);
2214 btrfs_set_super_log_root_level(fs_info
->super_copy
, 0);
2215 memcpy(fs_info
->super_for_commit
, fs_info
->super_copy
,
2216 sizeof(*fs_info
->super_copy
));
2218 btrfs_update_commit_device_size(fs_info
);
2219 btrfs_update_commit_device_bytes_used(fs_info
, cur_trans
);
2221 clear_bit(BTRFS_FS_LOG1_ERR
, &fs_info
->flags
);
2222 clear_bit(BTRFS_FS_LOG2_ERR
, &fs_info
->flags
);
2224 btrfs_trans_release_chunk_metadata(trans
);
2226 spin_lock(&fs_info
->trans_lock
);
2227 cur_trans
->state
= TRANS_STATE_UNBLOCKED
;
2228 fs_info
->running_transaction
= NULL
;
2229 spin_unlock(&fs_info
->trans_lock
);
2230 mutex_unlock(&fs_info
->reloc_mutex
);
2232 wake_up(&fs_info
->transaction_wait
);
2234 ret
= btrfs_write_and_wait_transaction(trans
, root
);
2236 btrfs_handle_fs_error(fs_info
, ret
,
2237 "Error while writing out transaction");
2238 mutex_unlock(&fs_info
->tree_log_mutex
);
2239 goto scrub_continue
;
2242 ret
= write_ctree_super(trans
, fs_info
, 0);
2244 mutex_unlock(&fs_info
->tree_log_mutex
);
2245 goto scrub_continue
;
2249 * the super is written, we can safely allow the tree-loggers
2250 * to go about their business
2252 mutex_unlock(&fs_info
->tree_log_mutex
);
2254 btrfs_finish_extent_commit(trans
, fs_info
);
2256 if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS
, &cur_trans
->flags
))
2257 btrfs_clear_space_info_full(fs_info
);
2259 fs_info
->last_trans_committed
= cur_trans
->transid
;
2261 * We needn't acquire the lock here because there is no other task
2262 * which can change it.
2264 cur_trans
->state
= TRANS_STATE_COMPLETED
;
2265 wake_up(&cur_trans
->commit_wait
);
2267 spin_lock(&fs_info
->trans_lock
);
2268 list_del_init(&cur_trans
->list
);
2269 spin_unlock(&fs_info
->trans_lock
);
2271 btrfs_put_transaction(cur_trans
);
2272 btrfs_put_transaction(cur_trans
);
2274 if (trans
->type
& __TRANS_FREEZABLE
)
2275 sb_end_intwrite(fs_info
->sb
);
2277 trace_btrfs_transaction_commit(root
);
2279 btrfs_scrub_continue(fs_info
);
2281 if (current
->journal_info
== trans
)
2282 current
->journal_info
= NULL
;
2284 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
2287 * If fs has been frozen, we can not handle delayed iputs, otherwise
2288 * it'll result in deadlock about SB_FREEZE_FS.
2290 if (current
!= fs_info
->transaction_kthread
&&
2291 current
!= fs_info
->cleaner_kthread
&& !fs_info
->fs_frozen
)
2292 btrfs_run_delayed_iputs(fs_info
);
2297 btrfs_scrub_continue(fs_info
);
2298 cleanup_transaction
:
2299 btrfs_trans_release_metadata(trans
, fs_info
);
2300 btrfs_trans_release_chunk_metadata(trans
);
2301 trans
->block_rsv
= NULL
;
2302 btrfs_warn(fs_info
, "Skipping commit of aborted transaction.");
2303 if (current
->journal_info
== trans
)
2304 current
->journal_info
= NULL
;
2305 cleanup_transaction(trans
, root
, ret
);
2311 * return < 0 if error
2312 * 0 if there are no more dead_roots at the time of call
2313 * 1 there are more to be processed, call me again
2315 * The return value indicates there are certainly more snapshots to delete, but
2316 * if there comes a new one during processing, it may return 0. We don't mind,
2317 * because btrfs_commit_super will poke cleaner thread and it will process it a
2318 * few seconds later.
2320 int btrfs_clean_one_deleted_snapshot(struct btrfs_root
*root
)
2323 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2325 spin_lock(&fs_info
->trans_lock
);
2326 if (list_empty(&fs_info
->dead_roots
)) {
2327 spin_unlock(&fs_info
->trans_lock
);
2330 root
= list_first_entry(&fs_info
->dead_roots
,
2331 struct btrfs_root
, root_list
);
2332 list_del_init(&root
->root_list
);
2333 spin_unlock(&fs_info
->trans_lock
);
2335 btrfs_debug(fs_info
, "cleaner removing %llu", root
->objectid
);
2337 btrfs_kill_all_delayed_nodes(root
);
2339 if (btrfs_header_backref_rev(root
->node
) <
2340 BTRFS_MIXED_BACKREF_REV
)
2341 ret
= btrfs_drop_snapshot(root
, NULL
, 0, 0);
2343 ret
= btrfs_drop_snapshot(root
, NULL
, 1, 0);
2345 return (ret
< 0) ? 0 : 1;
2348 void btrfs_apply_pending_changes(struct btrfs_fs_info
*fs_info
)
2353 prev
= xchg(&fs_info
->pending_changes
, 0);
2357 bit
= 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE
;
2359 btrfs_set_opt(fs_info
->mount_opt
, INODE_MAP_CACHE
);
2362 bit
= 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE
;
2364 btrfs_clear_opt(fs_info
->mount_opt
, INODE_MAP_CACHE
);
2367 bit
= 1 << BTRFS_PENDING_COMMIT
;
2369 btrfs_debug(fs_info
, "pending commit done");
2374 "unknown pending changes left 0x%lx, ignoring", prev
);