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>
27 #include "transaction.h"
30 #include "inode-map.h"
33 #define BTRFS_ROOT_TRANS_TAG 0
35 void put_transaction(struct btrfs_transaction
*transaction
)
37 WARN_ON(atomic_read(&transaction
->use_count
) == 0);
38 if (atomic_dec_and_test(&transaction
->use_count
)) {
39 BUG_ON(!list_empty(&transaction
->list
));
40 WARN_ON(transaction
->delayed_refs
.root
.rb_node
);
41 memset(transaction
, 0, sizeof(*transaction
));
42 kmem_cache_free(btrfs_transaction_cachep
, transaction
);
46 static noinline
void switch_commit_root(struct btrfs_root
*root
)
48 free_extent_buffer(root
->commit_root
);
49 root
->commit_root
= btrfs_root_node(root
);
53 * either allocate a new transaction or hop into the existing one
55 static noinline
int join_transaction(struct btrfs_root
*root
, int nofail
)
57 struct btrfs_transaction
*cur_trans
;
58 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
60 spin_lock(&fs_info
->trans_lock
);
62 /* The file system has been taken offline. No new transactions. */
63 if (fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
) {
64 spin_unlock(&fs_info
->trans_lock
);
68 if (fs_info
->trans_no_join
) {
70 spin_unlock(&fs_info
->trans_lock
);
75 cur_trans
= fs_info
->running_transaction
;
77 if (cur_trans
->aborted
) {
78 spin_unlock(&fs_info
->trans_lock
);
79 return cur_trans
->aborted
;
81 atomic_inc(&cur_trans
->use_count
);
82 atomic_inc(&cur_trans
->num_writers
);
83 cur_trans
->num_joined
++;
84 spin_unlock(&fs_info
->trans_lock
);
87 spin_unlock(&fs_info
->trans_lock
);
89 cur_trans
= kmem_cache_alloc(btrfs_transaction_cachep
, GFP_NOFS
);
93 spin_lock(&fs_info
->trans_lock
);
94 if (fs_info
->running_transaction
) {
96 * someone started a transaction after we unlocked. Make sure
97 * to redo the trans_no_join checks above
99 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
100 cur_trans
= fs_info
->running_transaction
;
102 } else if (fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
) {
103 spin_unlock(&fs_info
->trans_lock
);
104 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
108 atomic_set(&cur_trans
->num_writers
, 1);
109 cur_trans
->num_joined
= 0;
110 init_waitqueue_head(&cur_trans
->writer_wait
);
111 init_waitqueue_head(&cur_trans
->commit_wait
);
112 cur_trans
->in_commit
= 0;
113 cur_trans
->blocked
= 0;
115 * One for this trans handle, one so it will live on until we
116 * commit the transaction.
118 atomic_set(&cur_trans
->use_count
, 2);
119 cur_trans
->commit_done
= 0;
120 cur_trans
->start_time
= get_seconds();
122 cur_trans
->delayed_refs
.root
= RB_ROOT
;
123 cur_trans
->delayed_refs
.num_entries
= 0;
124 cur_trans
->delayed_refs
.num_heads_ready
= 0;
125 cur_trans
->delayed_refs
.num_heads
= 0;
126 cur_trans
->delayed_refs
.flushing
= 0;
127 cur_trans
->delayed_refs
.run_delayed_start
= 0;
130 * although the tree mod log is per file system and not per transaction,
131 * the log must never go across transaction boundaries.
134 if (!list_empty(&fs_info
->tree_mod_seq_list
)) {
135 printk(KERN_ERR
"btrfs: tree_mod_seq_list not empty when "
136 "creating a fresh transaction\n");
139 if (!RB_EMPTY_ROOT(&fs_info
->tree_mod_log
)) {
140 printk(KERN_ERR
"btrfs: tree_mod_log rb tree not empty when "
141 "creating a fresh transaction\n");
144 atomic_set(&fs_info
->tree_mod_seq
, 0);
146 spin_lock_init(&cur_trans
->commit_lock
);
147 spin_lock_init(&cur_trans
->delayed_refs
.lock
);
149 INIT_LIST_HEAD(&cur_trans
->pending_snapshots
);
150 list_add_tail(&cur_trans
->list
, &fs_info
->trans_list
);
151 extent_io_tree_init(&cur_trans
->dirty_pages
,
152 fs_info
->btree_inode
->i_mapping
);
153 fs_info
->generation
++;
154 cur_trans
->transid
= fs_info
->generation
;
155 fs_info
->running_transaction
= cur_trans
;
156 cur_trans
->aborted
= 0;
157 spin_unlock(&fs_info
->trans_lock
);
163 * this does all the record keeping required to make sure that a reference
164 * counted root is properly recorded in a given transaction. This is required
165 * to make sure the old root from before we joined the transaction is deleted
166 * when the transaction commits
168 static int record_root_in_trans(struct btrfs_trans_handle
*trans
,
169 struct btrfs_root
*root
)
171 if (root
->ref_cows
&& root
->last_trans
< trans
->transid
) {
172 WARN_ON(root
== root
->fs_info
->extent_root
);
173 WARN_ON(root
->commit_root
!= root
->node
);
176 * see below for in_trans_setup usage rules
177 * we have the reloc mutex held now, so there
178 * is only one writer in this function
180 root
->in_trans_setup
= 1;
182 /* make sure readers find in_trans_setup before
183 * they find our root->last_trans update
187 spin_lock(&root
->fs_info
->fs_roots_radix_lock
);
188 if (root
->last_trans
== trans
->transid
) {
189 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
192 radix_tree_tag_set(&root
->fs_info
->fs_roots_radix
,
193 (unsigned long)root
->root_key
.objectid
,
194 BTRFS_ROOT_TRANS_TAG
);
195 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
196 root
->last_trans
= trans
->transid
;
198 /* this is pretty tricky. We don't want to
199 * take the relocation lock in btrfs_record_root_in_trans
200 * unless we're really doing the first setup for this root in
203 * Normally we'd use root->last_trans as a flag to decide
204 * if we want to take the expensive mutex.
206 * But, we have to set root->last_trans before we
207 * init the relocation root, otherwise, we trip over warnings
208 * in ctree.c. The solution used here is to flag ourselves
209 * with root->in_trans_setup. When this is 1, we're still
210 * fixing up the reloc trees and everyone must wait.
212 * When this is zero, they can trust root->last_trans and fly
213 * through btrfs_record_root_in_trans without having to take the
214 * lock. smp_wmb() makes sure that all the writes above are
215 * done before we pop in the zero below
217 btrfs_init_reloc_root(trans
, root
);
219 root
->in_trans_setup
= 0;
225 int btrfs_record_root_in_trans(struct btrfs_trans_handle
*trans
,
226 struct btrfs_root
*root
)
232 * see record_root_in_trans for comments about in_trans_setup usage
236 if (root
->last_trans
== trans
->transid
&&
237 !root
->in_trans_setup
)
240 mutex_lock(&root
->fs_info
->reloc_mutex
);
241 record_root_in_trans(trans
, root
);
242 mutex_unlock(&root
->fs_info
->reloc_mutex
);
247 /* wait for commit against the current transaction to become unblocked
248 * when this is done, it is safe to start a new transaction, but the current
249 * transaction might not be fully on disk.
251 static void wait_current_trans(struct btrfs_root
*root
)
253 struct btrfs_transaction
*cur_trans
;
255 spin_lock(&root
->fs_info
->trans_lock
);
256 cur_trans
= root
->fs_info
->running_transaction
;
257 if (cur_trans
&& cur_trans
->blocked
) {
258 atomic_inc(&cur_trans
->use_count
);
259 spin_unlock(&root
->fs_info
->trans_lock
);
261 wait_event(root
->fs_info
->transaction_wait
,
262 !cur_trans
->blocked
);
263 put_transaction(cur_trans
);
265 spin_unlock(&root
->fs_info
->trans_lock
);
269 enum btrfs_trans_type
{
276 static int may_wait_transaction(struct btrfs_root
*root
, int type
)
278 if (root
->fs_info
->log_root_recovering
)
281 if (type
== TRANS_USERSPACE
)
284 if (type
== TRANS_START
&&
285 !atomic_read(&root
->fs_info
->open_ioctl_trans
))
291 static struct btrfs_trans_handle
*start_transaction(struct btrfs_root
*root
,
292 u64 num_items
, int type
)
294 struct btrfs_trans_handle
*h
;
295 struct btrfs_transaction
*cur_trans
;
298 u64 qgroup_reserved
= 0;
300 if (root
->fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
)
301 return ERR_PTR(-EROFS
);
303 if (current
->journal_info
) {
304 WARN_ON(type
!= TRANS_JOIN
&& type
!= TRANS_JOIN_NOLOCK
);
305 h
= current
->journal_info
;
307 h
->orig_rsv
= h
->block_rsv
;
313 * Do the reservation before we join the transaction so we can do all
314 * the appropriate flushing if need be.
316 if (num_items
> 0 && root
!= root
->fs_info
->chunk_root
) {
317 if (root
->fs_info
->quota_enabled
&&
318 is_fstree(root
->root_key
.objectid
)) {
319 qgroup_reserved
= num_items
* root
->leafsize
;
320 ret
= btrfs_qgroup_reserve(root
, qgroup_reserved
);
325 num_bytes
= btrfs_calc_trans_metadata_size(root
, num_items
);
326 ret
= btrfs_block_rsv_add(root
,
327 &root
->fs_info
->trans_block_rsv
,
333 h
= kmem_cache_alloc(btrfs_trans_handle_cachep
, GFP_NOFS
);
335 return ERR_PTR(-ENOMEM
);
337 if (may_wait_transaction(root
, type
))
338 wait_current_trans(root
);
341 ret
= join_transaction(root
, type
== TRANS_JOIN_NOLOCK
);
343 wait_current_trans(root
);
344 } while (ret
== -EBUSY
);
347 kmem_cache_free(btrfs_trans_handle_cachep
, h
);
351 cur_trans
= root
->fs_info
->running_transaction
;
353 h
->transid
= cur_trans
->transid
;
354 h
->transaction
= cur_trans
;
356 h
->bytes_reserved
= 0;
358 h
->delayed_ref_updates
= 0;
364 h
->qgroup_reserved
= qgroup_reserved
;
365 h
->delayed_ref_elem
.seq
= 0;
366 INIT_LIST_HEAD(&h
->qgroup_ref_list
);
369 if (cur_trans
->blocked
&& may_wait_transaction(root
, type
)) {
370 btrfs_commit_transaction(h
, root
);
375 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
376 h
->transid
, num_bytes
, 1);
377 h
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
378 h
->bytes_reserved
= num_bytes
;
382 btrfs_record_root_in_trans(h
, root
);
384 if (!current
->journal_info
&& type
!= TRANS_USERSPACE
)
385 current
->journal_info
= h
;
389 struct btrfs_trans_handle
*btrfs_start_transaction(struct btrfs_root
*root
,
392 return start_transaction(root
, num_items
, TRANS_START
);
394 struct btrfs_trans_handle
*btrfs_join_transaction(struct btrfs_root
*root
)
396 return start_transaction(root
, 0, TRANS_JOIN
);
399 struct btrfs_trans_handle
*btrfs_join_transaction_nolock(struct btrfs_root
*root
)
401 return start_transaction(root
, 0, TRANS_JOIN_NOLOCK
);
404 struct btrfs_trans_handle
*btrfs_start_ioctl_transaction(struct btrfs_root
*root
)
406 return start_transaction(root
, 0, TRANS_USERSPACE
);
409 /* wait for a transaction commit to be fully complete */
410 static noinline
void wait_for_commit(struct btrfs_root
*root
,
411 struct btrfs_transaction
*commit
)
413 wait_event(commit
->commit_wait
, commit
->commit_done
);
416 int btrfs_wait_for_commit(struct btrfs_root
*root
, u64 transid
)
418 struct btrfs_transaction
*cur_trans
= NULL
, *t
;
423 if (transid
<= root
->fs_info
->last_trans_committed
)
426 /* find specified transaction */
427 spin_lock(&root
->fs_info
->trans_lock
);
428 list_for_each_entry(t
, &root
->fs_info
->trans_list
, list
) {
429 if (t
->transid
== transid
) {
431 atomic_inc(&cur_trans
->use_count
);
434 if (t
->transid
> transid
)
437 spin_unlock(&root
->fs_info
->trans_lock
);
440 goto out
; /* bad transid */
442 /* find newest transaction that is committing | committed */
443 spin_lock(&root
->fs_info
->trans_lock
);
444 list_for_each_entry_reverse(t
, &root
->fs_info
->trans_list
,
450 atomic_inc(&cur_trans
->use_count
);
454 spin_unlock(&root
->fs_info
->trans_lock
);
456 goto out
; /* nothing committing|committed */
459 wait_for_commit(root
, cur_trans
);
461 put_transaction(cur_trans
);
467 void btrfs_throttle(struct btrfs_root
*root
)
469 if (!atomic_read(&root
->fs_info
->open_ioctl_trans
))
470 wait_current_trans(root
);
473 static int should_end_transaction(struct btrfs_trans_handle
*trans
,
474 struct btrfs_root
*root
)
478 ret
= btrfs_block_rsv_check(root
, &root
->fs_info
->global_block_rsv
, 5);
482 int btrfs_should_end_transaction(struct btrfs_trans_handle
*trans
,
483 struct btrfs_root
*root
)
485 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
490 if (cur_trans
->blocked
|| cur_trans
->delayed_refs
.flushing
)
493 updates
= trans
->delayed_ref_updates
;
494 trans
->delayed_ref_updates
= 0;
496 err
= btrfs_run_delayed_refs(trans
, root
, updates
);
497 if (err
) /* Error code will also eval true */
501 return should_end_transaction(trans
, root
);
504 static int __btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
505 struct btrfs_root
*root
, int throttle
, int lock
)
507 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
508 struct btrfs_fs_info
*info
= root
->fs_info
;
512 if (--trans
->use_count
) {
513 trans
->block_rsv
= trans
->orig_rsv
;
518 * do the qgroup accounting as early as possible
520 err
= btrfs_delayed_refs_qgroup_accounting(trans
, info
);
522 btrfs_trans_release_metadata(trans
, root
);
523 trans
->block_rsv
= NULL
;
525 * the same root has to be passed to start_transaction and
526 * end_transaction. Subvolume quota depends on this.
528 WARN_ON(trans
->root
!= root
);
530 if (trans
->qgroup_reserved
) {
531 btrfs_qgroup_free(root
, trans
->qgroup_reserved
);
532 trans
->qgroup_reserved
= 0;
536 unsigned long cur
= trans
->delayed_ref_updates
;
537 trans
->delayed_ref_updates
= 0;
539 trans
->transaction
->delayed_refs
.num_heads_ready
> 64) {
540 trans
->delayed_ref_updates
= 0;
541 btrfs_run_delayed_refs(trans
, root
, cur
);
547 btrfs_trans_release_metadata(trans
, root
);
548 trans
->block_rsv
= NULL
;
550 if (lock
&& !atomic_read(&root
->fs_info
->open_ioctl_trans
) &&
551 should_end_transaction(trans
, root
)) {
552 trans
->transaction
->blocked
= 1;
556 if (lock
&& cur_trans
->blocked
&& !cur_trans
->in_commit
) {
559 * We may race with somebody else here so end up having
560 * to call end_transaction on ourselves again, so inc
564 return btrfs_commit_transaction(trans
, root
);
566 wake_up_process(info
->transaction_kthread
);
570 WARN_ON(cur_trans
!= info
->running_transaction
);
571 WARN_ON(atomic_read(&cur_trans
->num_writers
) < 1);
572 atomic_dec(&cur_trans
->num_writers
);
575 if (waitqueue_active(&cur_trans
->writer_wait
))
576 wake_up(&cur_trans
->writer_wait
);
577 put_transaction(cur_trans
);
579 if (current
->journal_info
== trans
)
580 current
->journal_info
= NULL
;
583 btrfs_run_delayed_iputs(root
);
585 if (trans
->aborted
||
586 root
->fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
) {
589 assert_qgroups_uptodate(trans
);
591 memset(trans
, 0, sizeof(*trans
));
592 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
596 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
597 struct btrfs_root
*root
)
601 ret
= __btrfs_end_transaction(trans
, root
, 0, 1);
607 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
,
608 struct btrfs_root
*root
)
612 ret
= __btrfs_end_transaction(trans
, root
, 1, 1);
618 int btrfs_end_transaction_nolock(struct btrfs_trans_handle
*trans
,
619 struct btrfs_root
*root
)
623 ret
= __btrfs_end_transaction(trans
, root
, 0, 0);
629 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle
*trans
,
630 struct btrfs_root
*root
)
632 return __btrfs_end_transaction(trans
, root
, 1, 1);
636 * when btree blocks are allocated, they have some corresponding bits set for
637 * them in one of two extent_io trees. This is used to make sure all of
638 * those extents are sent to disk but does not wait on them
640 int btrfs_write_marked_extents(struct btrfs_root
*root
,
641 struct extent_io_tree
*dirty_pages
, int mark
)
645 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
649 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
651 convert_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
, mark
,
653 err
= filemap_fdatawrite_range(mapping
, start
, end
);
665 * when btree blocks are allocated, they have some corresponding bits set for
666 * them in one of two extent_io trees. This is used to make sure all of
667 * those extents are on disk for transaction or log commit. We wait
668 * on all the pages and clear them from the dirty pages state tree
670 int btrfs_wait_marked_extents(struct btrfs_root
*root
,
671 struct extent_io_tree
*dirty_pages
, int mark
)
675 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
679 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
681 clear_extent_bits(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
, GFP_NOFS
);
682 err
= filemap_fdatawait_range(mapping
, start
, end
);
694 * when btree blocks are allocated, they have some corresponding bits set for
695 * them in one of two extent_io trees. This is used to make sure all of
696 * those extents are on disk for transaction or log commit
698 int btrfs_write_and_wait_marked_extents(struct btrfs_root
*root
,
699 struct extent_io_tree
*dirty_pages
, int mark
)
704 ret
= btrfs_write_marked_extents(root
, dirty_pages
, mark
);
705 ret2
= btrfs_wait_marked_extents(root
, dirty_pages
, mark
);
714 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
,
715 struct btrfs_root
*root
)
717 if (!trans
|| !trans
->transaction
) {
718 struct inode
*btree_inode
;
719 btree_inode
= root
->fs_info
->btree_inode
;
720 return filemap_write_and_wait(btree_inode
->i_mapping
);
722 return btrfs_write_and_wait_marked_extents(root
,
723 &trans
->transaction
->dirty_pages
,
728 * this is used to update the root pointer in the tree of tree roots.
730 * But, in the case of the extent allocation tree, updating the root
731 * pointer may allocate blocks which may change the root of the extent
734 * So, this loops and repeats and makes sure the cowonly root didn't
735 * change while the root pointer was being updated in the metadata.
737 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
738 struct btrfs_root
*root
)
743 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
745 old_root_used
= btrfs_root_used(&root
->root_item
);
746 btrfs_write_dirty_block_groups(trans
, root
);
749 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
750 if (old_root_bytenr
== root
->node
->start
&&
751 old_root_used
== btrfs_root_used(&root
->root_item
))
754 btrfs_set_root_node(&root
->root_item
, root
->node
);
755 ret
= btrfs_update_root(trans
, tree_root
,
761 old_root_used
= btrfs_root_used(&root
->root_item
);
762 ret
= btrfs_write_dirty_block_groups(trans
, root
);
767 if (root
!= root
->fs_info
->extent_root
)
768 switch_commit_root(root
);
774 * update all the cowonly tree roots on disk
776 * The error handling in this function may not be obvious. Any of the
777 * failures will cause the file system to go offline. We still need
778 * to clean up the delayed refs.
780 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
,
781 struct btrfs_root
*root
)
783 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
784 struct list_head
*next
;
785 struct extent_buffer
*eb
;
788 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
792 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
793 ret
= btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
,
795 btrfs_tree_unlock(eb
);
796 free_extent_buffer(eb
);
801 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
805 ret
= btrfs_run_dev_stats(trans
, root
->fs_info
);
808 ret
= btrfs_run_qgroups(trans
, root
->fs_info
);
811 /* run_qgroups might have added some more refs */
812 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
815 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
816 next
= fs_info
->dirty_cowonly_roots
.next
;
818 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
820 ret
= update_cowonly_root(trans
, root
);
825 down_write(&fs_info
->extent_commit_sem
);
826 switch_commit_root(fs_info
->extent_root
);
827 up_write(&fs_info
->extent_commit_sem
);
833 * dead roots are old snapshots that need to be deleted. This allocates
834 * a dirty root struct and adds it into the list of dead roots that need to
837 int btrfs_add_dead_root(struct btrfs_root
*root
)
839 spin_lock(&root
->fs_info
->trans_lock
);
840 list_add(&root
->root_list
, &root
->fs_info
->dead_roots
);
841 spin_unlock(&root
->fs_info
->trans_lock
);
846 * update all the cowonly tree roots on disk
848 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
,
849 struct btrfs_root
*root
)
851 struct btrfs_root
*gang
[8];
852 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
857 spin_lock(&fs_info
->fs_roots_radix_lock
);
859 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
862 BTRFS_ROOT_TRANS_TAG
);
865 for (i
= 0; i
< ret
; i
++) {
867 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
868 (unsigned long)root
->root_key
.objectid
,
869 BTRFS_ROOT_TRANS_TAG
);
870 spin_unlock(&fs_info
->fs_roots_radix_lock
);
872 btrfs_free_log(trans
, root
);
873 btrfs_update_reloc_root(trans
, root
);
874 btrfs_orphan_commit_root(trans
, root
);
876 btrfs_save_ino_cache(root
, trans
);
878 /* see comments in should_cow_block() */
882 if (root
->commit_root
!= root
->node
) {
883 mutex_lock(&root
->fs_commit_mutex
);
884 switch_commit_root(root
);
885 btrfs_unpin_free_ino(root
);
886 mutex_unlock(&root
->fs_commit_mutex
);
888 btrfs_set_root_node(&root
->root_item
,
892 err
= btrfs_update_root(trans
, fs_info
->tree_root
,
895 spin_lock(&fs_info
->fs_roots_radix_lock
);
900 spin_unlock(&fs_info
->fs_roots_radix_lock
);
905 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
906 * otherwise every leaf in the btree is read and defragged.
908 int btrfs_defrag_root(struct btrfs_root
*root
, int cacheonly
)
910 struct btrfs_fs_info
*info
= root
->fs_info
;
911 struct btrfs_trans_handle
*trans
;
915 if (xchg(&root
->defrag_running
, 1))
919 trans
= btrfs_start_transaction(root
, 0);
921 return PTR_ERR(trans
);
923 ret
= btrfs_defrag_leaves(trans
, root
, cacheonly
);
925 nr
= trans
->blocks_used
;
926 btrfs_end_transaction(trans
, root
);
927 btrfs_btree_balance_dirty(info
->tree_root
, nr
);
930 if (btrfs_fs_closing(root
->fs_info
) || ret
!= -EAGAIN
)
933 root
->defrag_running
= 0;
938 * new snapshots need to be created at a very specific time in the
939 * transaction commit. This does the actual creation
941 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
942 struct btrfs_fs_info
*fs_info
,
943 struct btrfs_pending_snapshot
*pending
)
945 struct btrfs_key key
;
946 struct btrfs_root_item
*new_root_item
;
947 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
948 struct btrfs_root
*root
= pending
->root
;
949 struct btrfs_root
*parent_root
;
950 struct btrfs_block_rsv
*rsv
;
951 struct inode
*parent_inode
;
952 struct dentry
*parent
;
953 struct dentry
*dentry
;
954 struct extent_buffer
*tmp
;
955 struct extent_buffer
*old
;
962 rsv
= trans
->block_rsv
;
964 new_root_item
= kmalloc(sizeof(*new_root_item
), GFP_NOFS
);
965 if (!new_root_item
) {
966 ret
= pending
->error
= -ENOMEM
;
970 ret
= btrfs_find_free_objectid(tree_root
, &objectid
);
972 pending
->error
= ret
;
976 btrfs_reloc_pre_snapshot(trans
, pending
, &to_reserve
);
978 if (to_reserve
> 0) {
979 ret
= btrfs_block_rsv_add_noflush(root
, &pending
->block_rsv
,
982 pending
->error
= ret
;
987 ret
= btrfs_qgroup_inherit(trans
, fs_info
, root
->root_key
.objectid
,
988 objectid
, pending
->inherit
);
989 kfree(pending
->inherit
);
991 pending
->error
= ret
;
995 key
.objectid
= objectid
;
996 key
.offset
= (u64
)-1;
997 key
.type
= BTRFS_ROOT_ITEM_KEY
;
999 trans
->block_rsv
= &pending
->block_rsv
;
1001 dentry
= pending
->dentry
;
1002 parent
= dget_parent(dentry
);
1003 parent_inode
= parent
->d_inode
;
1004 parent_root
= BTRFS_I(parent_inode
)->root
;
1005 record_root_in_trans(trans
, parent_root
);
1008 * insert the directory item
1010 ret
= btrfs_set_inode_index(parent_inode
, &index
);
1011 BUG_ON(ret
); /* -ENOMEM */
1012 ret
= btrfs_insert_dir_item(trans
, parent_root
,
1013 dentry
->d_name
.name
, dentry
->d_name
.len
,
1015 BTRFS_FT_DIR
, index
);
1016 if (ret
== -EEXIST
) {
1017 pending
->error
= -EEXIST
;
1021 goto abort_trans_dput
;
1024 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
1025 dentry
->d_name
.len
* 2);
1026 ret
= btrfs_update_inode(trans
, parent_root
, parent_inode
);
1028 goto abort_trans_dput
;
1031 * pull in the delayed directory update
1032 * and the delayed inode item
1033 * otherwise we corrupt the FS during
1036 ret
= btrfs_run_delayed_items(trans
, root
);
1037 if (ret
) { /* Transaction aborted */
1042 record_root_in_trans(trans
, root
);
1043 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
1044 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
1045 btrfs_check_and_init_root_item(new_root_item
);
1047 root_flags
= btrfs_root_flags(new_root_item
);
1048 if (pending
->readonly
)
1049 root_flags
|= BTRFS_ROOT_SUBVOL_RDONLY
;
1051 root_flags
&= ~BTRFS_ROOT_SUBVOL_RDONLY
;
1052 btrfs_set_root_flags(new_root_item
, root_flags
);
1054 old
= btrfs_lock_root_node(root
);
1055 ret
= btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
);
1057 btrfs_tree_unlock(old
);
1058 free_extent_buffer(old
);
1059 goto abort_trans_dput
;
1062 btrfs_set_lock_blocking(old
);
1064 ret
= btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
1065 /* clean up in any case */
1066 btrfs_tree_unlock(old
);
1067 free_extent_buffer(old
);
1069 goto abort_trans_dput
;
1071 /* see comments in should_cow_block() */
1072 root
->force_cow
= 1;
1075 btrfs_set_root_node(new_root_item
, tmp
);
1076 /* record when the snapshot was created in key.offset */
1077 key
.offset
= trans
->transid
;
1078 ret
= btrfs_insert_root(trans
, tree_root
, &key
, new_root_item
);
1079 btrfs_tree_unlock(tmp
);
1080 free_extent_buffer(tmp
);
1082 goto abort_trans_dput
;
1085 * insert root back/forward references
1087 ret
= btrfs_add_root_ref(trans
, tree_root
, objectid
,
1088 parent_root
->root_key
.objectid
,
1089 btrfs_ino(parent_inode
), index
,
1090 dentry
->d_name
.name
, dentry
->d_name
.len
);
1095 key
.offset
= (u64
)-1;
1096 pending
->snap
= btrfs_read_fs_root_no_name(root
->fs_info
, &key
);
1097 if (IS_ERR(pending
->snap
)) {
1098 ret
= PTR_ERR(pending
->snap
);
1102 ret
= btrfs_reloc_post_snapshot(trans
, pending
);
1107 kfree(new_root_item
);
1108 trans
->block_rsv
= rsv
;
1109 btrfs_block_rsv_release(root
, &pending
->block_rsv
, (u64
)-1);
1115 btrfs_abort_transaction(trans
, root
, ret
);
1120 * create all the snapshots we've scheduled for creation
1122 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
,
1123 struct btrfs_fs_info
*fs_info
)
1125 struct btrfs_pending_snapshot
*pending
;
1126 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
1128 list_for_each_entry(pending
, head
, list
)
1129 create_pending_snapshot(trans
, fs_info
, pending
);
1133 static void update_super_roots(struct btrfs_root
*root
)
1135 struct btrfs_root_item
*root_item
;
1136 struct btrfs_super_block
*super
;
1138 super
= root
->fs_info
->super_copy
;
1140 root_item
= &root
->fs_info
->chunk_root
->root_item
;
1141 super
->chunk_root
= root_item
->bytenr
;
1142 super
->chunk_root_generation
= root_item
->generation
;
1143 super
->chunk_root_level
= root_item
->level
;
1145 root_item
= &root
->fs_info
->tree_root
->root_item
;
1146 super
->root
= root_item
->bytenr
;
1147 super
->generation
= root_item
->generation
;
1148 super
->root_level
= root_item
->level
;
1149 if (btrfs_test_opt(root
, SPACE_CACHE
))
1150 super
->cache_generation
= root_item
->generation
;
1153 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
1156 spin_lock(&info
->trans_lock
);
1157 if (info
->running_transaction
)
1158 ret
= info
->running_transaction
->in_commit
;
1159 spin_unlock(&info
->trans_lock
);
1163 int btrfs_transaction_blocked(struct btrfs_fs_info
*info
)
1166 spin_lock(&info
->trans_lock
);
1167 if (info
->running_transaction
)
1168 ret
= info
->running_transaction
->blocked
;
1169 spin_unlock(&info
->trans_lock
);
1174 * wait for the current transaction commit to start and block subsequent
1177 static void wait_current_trans_commit_start(struct btrfs_root
*root
,
1178 struct btrfs_transaction
*trans
)
1180 wait_event(root
->fs_info
->transaction_blocked_wait
, trans
->in_commit
);
1184 * wait for the current transaction to start and then become unblocked.
1187 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root
*root
,
1188 struct btrfs_transaction
*trans
)
1190 wait_event(root
->fs_info
->transaction_wait
,
1191 trans
->commit_done
|| (trans
->in_commit
&& !trans
->blocked
));
1195 * commit transactions asynchronously. once btrfs_commit_transaction_async
1196 * returns, any subsequent transaction will not be allowed to join.
1198 struct btrfs_async_commit
{
1199 struct btrfs_trans_handle
*newtrans
;
1200 struct btrfs_root
*root
;
1201 struct delayed_work work
;
1204 static void do_async_commit(struct work_struct
*work
)
1206 struct btrfs_async_commit
*ac
=
1207 container_of(work
, struct btrfs_async_commit
, work
.work
);
1209 btrfs_commit_transaction(ac
->newtrans
, ac
->root
);
1213 int btrfs_commit_transaction_async(struct btrfs_trans_handle
*trans
,
1214 struct btrfs_root
*root
,
1215 int wait_for_unblock
)
1217 struct btrfs_async_commit
*ac
;
1218 struct btrfs_transaction
*cur_trans
;
1220 ac
= kmalloc(sizeof(*ac
), GFP_NOFS
);
1224 INIT_DELAYED_WORK(&ac
->work
, do_async_commit
);
1226 ac
->newtrans
= btrfs_join_transaction(root
);
1227 if (IS_ERR(ac
->newtrans
)) {
1228 int err
= PTR_ERR(ac
->newtrans
);
1233 /* take transaction reference */
1234 cur_trans
= trans
->transaction
;
1235 atomic_inc(&cur_trans
->use_count
);
1237 btrfs_end_transaction(trans
, root
);
1238 schedule_delayed_work(&ac
->work
, 0);
1240 /* wait for transaction to start and unblock */
1241 if (wait_for_unblock
)
1242 wait_current_trans_commit_start_and_unblock(root
, cur_trans
);
1244 wait_current_trans_commit_start(root
, cur_trans
);
1246 if (current
->journal_info
== trans
)
1247 current
->journal_info
= NULL
;
1249 put_transaction(cur_trans
);
1254 static void cleanup_transaction(struct btrfs_trans_handle
*trans
,
1255 struct btrfs_root
*root
, int err
)
1257 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1259 WARN_ON(trans
->use_count
> 1);
1261 btrfs_abort_transaction(trans
, root
, err
);
1263 spin_lock(&root
->fs_info
->trans_lock
);
1264 list_del_init(&cur_trans
->list
);
1265 if (cur_trans
== root
->fs_info
->running_transaction
) {
1266 root
->fs_info
->running_transaction
= NULL
;
1267 root
->fs_info
->trans_no_join
= 0;
1269 spin_unlock(&root
->fs_info
->trans_lock
);
1271 btrfs_cleanup_one_transaction(trans
->transaction
, root
);
1273 put_transaction(cur_trans
);
1274 put_transaction(cur_trans
);
1276 trace_btrfs_transaction_commit(root
);
1278 btrfs_scrub_continue(root
);
1280 if (current
->journal_info
== trans
)
1281 current
->journal_info
= NULL
;
1283 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1287 * btrfs_transaction state sequence:
1288 * in_commit = 0, blocked = 0 (initial)
1289 * in_commit = 1, blocked = 1
1293 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
,
1294 struct btrfs_root
*root
)
1296 unsigned long joined
= 0;
1297 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1298 struct btrfs_transaction
*prev_trans
= NULL
;
1301 int should_grow
= 0;
1302 unsigned long now
= get_seconds();
1303 int flush_on_commit
= btrfs_test_opt(root
, FLUSHONCOMMIT
);
1305 btrfs_run_ordered_operations(root
, 0);
1307 if (cur_trans
->aborted
)
1308 goto cleanup_transaction
;
1310 /* make a pass through all the delayed refs we have so far
1311 * any runnings procs may add more while we are here
1313 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1315 goto cleanup_transaction
;
1317 btrfs_trans_release_metadata(trans
, root
);
1318 trans
->block_rsv
= NULL
;
1320 cur_trans
= trans
->transaction
;
1323 * set the flushing flag so procs in this transaction have to
1324 * start sending their work down.
1326 cur_trans
->delayed_refs
.flushing
= 1;
1328 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1330 goto cleanup_transaction
;
1332 spin_lock(&cur_trans
->commit_lock
);
1333 if (cur_trans
->in_commit
) {
1334 spin_unlock(&cur_trans
->commit_lock
);
1335 atomic_inc(&cur_trans
->use_count
);
1336 ret
= btrfs_end_transaction(trans
, root
);
1338 wait_for_commit(root
, cur_trans
);
1340 put_transaction(cur_trans
);
1345 trans
->transaction
->in_commit
= 1;
1346 trans
->transaction
->blocked
= 1;
1347 spin_unlock(&cur_trans
->commit_lock
);
1348 wake_up(&root
->fs_info
->transaction_blocked_wait
);
1350 spin_lock(&root
->fs_info
->trans_lock
);
1351 if (cur_trans
->list
.prev
!= &root
->fs_info
->trans_list
) {
1352 prev_trans
= list_entry(cur_trans
->list
.prev
,
1353 struct btrfs_transaction
, list
);
1354 if (!prev_trans
->commit_done
) {
1355 atomic_inc(&prev_trans
->use_count
);
1356 spin_unlock(&root
->fs_info
->trans_lock
);
1358 wait_for_commit(root
, prev_trans
);
1360 put_transaction(prev_trans
);
1362 spin_unlock(&root
->fs_info
->trans_lock
);
1365 spin_unlock(&root
->fs_info
->trans_lock
);
1368 if (!btrfs_test_opt(root
, SSD
) &&
1369 (now
< cur_trans
->start_time
|| now
- cur_trans
->start_time
< 1))
1373 int snap_pending
= 0;
1375 joined
= cur_trans
->num_joined
;
1376 if (!list_empty(&trans
->transaction
->pending_snapshots
))
1379 WARN_ON(cur_trans
!= trans
->transaction
);
1381 if (flush_on_commit
|| snap_pending
) {
1382 btrfs_start_delalloc_inodes(root
, 1);
1383 btrfs_wait_ordered_extents(root
, 0, 1);
1386 ret
= btrfs_run_delayed_items(trans
, root
);
1388 goto cleanup_transaction
;
1391 * running the delayed items may have added new refs. account
1392 * them now so that they hinder processing of more delayed refs
1393 * as little as possible.
1395 btrfs_delayed_refs_qgroup_accounting(trans
, root
->fs_info
);
1398 * rename don't use btrfs_join_transaction, so, once we
1399 * set the transaction to blocked above, we aren't going
1400 * to get any new ordered operations. We can safely run
1401 * it here and no for sure that nothing new will be added
1404 btrfs_run_ordered_operations(root
, 1);
1406 prepare_to_wait(&cur_trans
->writer_wait
, &wait
,
1407 TASK_UNINTERRUPTIBLE
);
1409 if (atomic_read(&cur_trans
->num_writers
) > 1)
1410 schedule_timeout(MAX_SCHEDULE_TIMEOUT
);
1411 else if (should_grow
)
1412 schedule_timeout(1);
1414 finish_wait(&cur_trans
->writer_wait
, &wait
);
1415 } while (atomic_read(&cur_trans
->num_writers
) > 1 ||
1416 (should_grow
&& cur_trans
->num_joined
!= joined
));
1419 * Ok now we need to make sure to block out any other joins while we
1420 * commit the transaction. We could have started a join before setting
1421 * no_join so make sure to wait for num_writers to == 1 again.
1423 spin_lock(&root
->fs_info
->trans_lock
);
1424 root
->fs_info
->trans_no_join
= 1;
1425 spin_unlock(&root
->fs_info
->trans_lock
);
1426 wait_event(cur_trans
->writer_wait
,
1427 atomic_read(&cur_trans
->num_writers
) == 1);
1430 * the reloc mutex makes sure that we stop
1431 * the balancing code from coming in and moving
1432 * extents around in the middle of the commit
1434 mutex_lock(&root
->fs_info
->reloc_mutex
);
1436 ret
= btrfs_run_delayed_items(trans
, root
);
1438 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1439 goto cleanup_transaction
;
1442 ret
= create_pending_snapshots(trans
, root
->fs_info
);
1444 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1445 goto cleanup_transaction
;
1448 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1450 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1451 goto cleanup_transaction
;
1455 * make sure none of the code above managed to slip in a
1458 btrfs_assert_delayed_root_empty(root
);
1460 WARN_ON(cur_trans
!= trans
->transaction
);
1462 btrfs_scrub_pause(root
);
1463 /* btrfs_commit_tree_roots is responsible for getting the
1464 * various roots consistent with each other. Every pointer
1465 * in the tree of tree roots has to point to the most up to date
1466 * root for every subvolume and other tree. So, we have to keep
1467 * the tree logging code from jumping in and changing any
1470 * At this point in the commit, there can't be any tree-log
1471 * writers, but a little lower down we drop the trans mutex
1472 * and let new people in. By holding the tree_log_mutex
1473 * from now until after the super is written, we avoid races
1474 * with the tree-log code.
1476 mutex_lock(&root
->fs_info
->tree_log_mutex
);
1478 ret
= commit_fs_roots(trans
, root
);
1480 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1481 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1482 goto cleanup_transaction
;
1485 /* commit_fs_roots gets rid of all the tree log roots, it is now
1486 * safe to free the root of tree log roots
1488 btrfs_free_log_root_tree(trans
, root
->fs_info
);
1490 ret
= commit_cowonly_roots(trans
, root
);
1492 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1493 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1494 goto cleanup_transaction
;
1497 btrfs_prepare_extent_commit(trans
, root
);
1499 cur_trans
= root
->fs_info
->running_transaction
;
1501 btrfs_set_root_node(&root
->fs_info
->tree_root
->root_item
,
1502 root
->fs_info
->tree_root
->node
);
1503 switch_commit_root(root
->fs_info
->tree_root
);
1505 btrfs_set_root_node(&root
->fs_info
->chunk_root
->root_item
,
1506 root
->fs_info
->chunk_root
->node
);
1507 switch_commit_root(root
->fs_info
->chunk_root
);
1509 assert_qgroups_uptodate(trans
);
1510 update_super_roots(root
);
1512 if (!root
->fs_info
->log_root_recovering
) {
1513 btrfs_set_super_log_root(root
->fs_info
->super_copy
, 0);
1514 btrfs_set_super_log_root_level(root
->fs_info
->super_copy
, 0);
1517 memcpy(root
->fs_info
->super_for_commit
, root
->fs_info
->super_copy
,
1518 sizeof(*root
->fs_info
->super_copy
));
1520 trans
->transaction
->blocked
= 0;
1521 spin_lock(&root
->fs_info
->trans_lock
);
1522 root
->fs_info
->running_transaction
= NULL
;
1523 root
->fs_info
->trans_no_join
= 0;
1524 spin_unlock(&root
->fs_info
->trans_lock
);
1525 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1527 wake_up(&root
->fs_info
->transaction_wait
);
1529 ret
= btrfs_write_and_wait_transaction(trans
, root
);
1531 btrfs_error(root
->fs_info
, ret
,
1532 "Error while writing out transaction.");
1533 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1534 goto cleanup_transaction
;
1537 ret
= write_ctree_super(trans
, root
, 0);
1539 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1540 goto cleanup_transaction
;
1544 * the super is written, we can safely allow the tree-loggers
1545 * to go about their business
1547 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1549 btrfs_finish_extent_commit(trans
, root
);
1551 cur_trans
->commit_done
= 1;
1553 root
->fs_info
->last_trans_committed
= cur_trans
->transid
;
1555 wake_up(&cur_trans
->commit_wait
);
1557 spin_lock(&root
->fs_info
->trans_lock
);
1558 list_del_init(&cur_trans
->list
);
1559 spin_unlock(&root
->fs_info
->trans_lock
);
1561 put_transaction(cur_trans
);
1562 put_transaction(cur_trans
);
1564 trace_btrfs_transaction_commit(root
);
1566 btrfs_scrub_continue(root
);
1568 if (current
->journal_info
== trans
)
1569 current
->journal_info
= NULL
;
1571 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1573 if (current
!= root
->fs_info
->transaction_kthread
)
1574 btrfs_run_delayed_iputs(root
);
1578 cleanup_transaction
:
1579 btrfs_trans_release_metadata(trans
, root
);
1580 trans
->block_rsv
= NULL
;
1581 btrfs_printk(root
->fs_info
, "Skipping commit of aborted transaction.\n");
1583 if (current
->journal_info
== trans
)
1584 current
->journal_info
= NULL
;
1585 cleanup_transaction(trans
, root
, ret
);
1591 * interface function to delete all the snapshots we have scheduled for deletion
1593 int btrfs_clean_old_snapshots(struct btrfs_root
*root
)
1596 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1598 spin_lock(&fs_info
->trans_lock
);
1599 list_splice_init(&fs_info
->dead_roots
, &list
);
1600 spin_unlock(&fs_info
->trans_lock
);
1602 while (!list_empty(&list
)) {
1605 root
= list_entry(list
.next
, struct btrfs_root
, root_list
);
1606 list_del(&root
->root_list
);
1608 btrfs_kill_all_delayed_nodes(root
);
1610 if (btrfs_header_backref_rev(root
->node
) <
1611 BTRFS_MIXED_BACKREF_REV
)
1612 ret
= btrfs_drop_snapshot(root
, NULL
, 0, 0);
1614 ret
=btrfs_drop_snapshot(root
, NULL
, 1, 0);