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 (root
->fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
) {
103 spin_unlock(&root
->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
;
299 if (root
->fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
)
300 return ERR_PTR(-EROFS
);
302 if (current
->journal_info
) {
303 WARN_ON(type
!= TRANS_JOIN
&& type
!= TRANS_JOIN_NOLOCK
);
304 h
= current
->journal_info
;
306 h
->orig_rsv
= h
->block_rsv
;
312 * Do the reservation before we join the transaction so we can do all
313 * the appropriate flushing if need be.
315 if (num_items
> 0 && root
!= root
->fs_info
->chunk_root
) {
316 num_bytes
= btrfs_calc_trans_metadata_size(root
, num_items
);
317 ret
= btrfs_block_rsv_add(root
,
318 &root
->fs_info
->trans_block_rsv
,
324 h
= kmem_cache_alloc(btrfs_trans_handle_cachep
, GFP_NOFS
);
326 return ERR_PTR(-ENOMEM
);
328 if (may_wait_transaction(root
, type
))
329 wait_current_trans(root
);
332 ret
= join_transaction(root
, type
== TRANS_JOIN_NOLOCK
);
334 wait_current_trans(root
);
335 } while (ret
== -EBUSY
);
338 kmem_cache_free(btrfs_trans_handle_cachep
, h
);
342 cur_trans
= root
->fs_info
->running_transaction
;
344 h
->transid
= cur_trans
->transid
;
345 h
->transaction
= cur_trans
;
347 h
->bytes_reserved
= 0;
349 h
->delayed_ref_updates
= 0;
354 h
->delayed_ref_elem
.seq
= 0;
355 INIT_LIST_HEAD(&h
->qgroup_ref_list
);
358 if (cur_trans
->blocked
&& may_wait_transaction(root
, type
)) {
359 btrfs_commit_transaction(h
, root
);
364 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
365 h
->transid
, num_bytes
, 1);
366 h
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
367 h
->bytes_reserved
= num_bytes
;
371 btrfs_record_root_in_trans(h
, root
);
373 if (!current
->journal_info
&& type
!= TRANS_USERSPACE
)
374 current
->journal_info
= h
;
378 struct btrfs_trans_handle
*btrfs_start_transaction(struct btrfs_root
*root
,
381 return start_transaction(root
, num_items
, TRANS_START
);
383 struct btrfs_trans_handle
*btrfs_join_transaction(struct btrfs_root
*root
)
385 return start_transaction(root
, 0, TRANS_JOIN
);
388 struct btrfs_trans_handle
*btrfs_join_transaction_nolock(struct btrfs_root
*root
)
390 return start_transaction(root
, 0, TRANS_JOIN_NOLOCK
);
393 struct btrfs_trans_handle
*btrfs_start_ioctl_transaction(struct btrfs_root
*root
)
395 return start_transaction(root
, 0, TRANS_USERSPACE
);
398 /* wait for a transaction commit to be fully complete */
399 static noinline
void wait_for_commit(struct btrfs_root
*root
,
400 struct btrfs_transaction
*commit
)
402 wait_event(commit
->commit_wait
, commit
->commit_done
);
405 int btrfs_wait_for_commit(struct btrfs_root
*root
, u64 transid
)
407 struct btrfs_transaction
*cur_trans
= NULL
, *t
;
412 if (transid
<= root
->fs_info
->last_trans_committed
)
415 /* find specified transaction */
416 spin_lock(&root
->fs_info
->trans_lock
);
417 list_for_each_entry(t
, &root
->fs_info
->trans_list
, list
) {
418 if (t
->transid
== transid
) {
420 atomic_inc(&cur_trans
->use_count
);
423 if (t
->transid
> transid
)
426 spin_unlock(&root
->fs_info
->trans_lock
);
429 goto out
; /* bad transid */
431 /* find newest transaction that is committing | committed */
432 spin_lock(&root
->fs_info
->trans_lock
);
433 list_for_each_entry_reverse(t
, &root
->fs_info
->trans_list
,
439 atomic_inc(&cur_trans
->use_count
);
443 spin_unlock(&root
->fs_info
->trans_lock
);
445 goto out
; /* nothing committing|committed */
448 wait_for_commit(root
, cur_trans
);
450 put_transaction(cur_trans
);
456 void btrfs_throttle(struct btrfs_root
*root
)
458 if (!atomic_read(&root
->fs_info
->open_ioctl_trans
))
459 wait_current_trans(root
);
462 static int should_end_transaction(struct btrfs_trans_handle
*trans
,
463 struct btrfs_root
*root
)
467 ret
= btrfs_block_rsv_check(root
, &root
->fs_info
->global_block_rsv
, 5);
471 int btrfs_should_end_transaction(struct btrfs_trans_handle
*trans
,
472 struct btrfs_root
*root
)
474 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
475 struct btrfs_block_rsv
*rsv
= trans
->block_rsv
;
480 if (cur_trans
->blocked
|| cur_trans
->delayed_refs
.flushing
)
484 * We need to do this in case we're deleting csums so the global block
485 * rsv get's used instead of the csum block rsv.
487 trans
->block_rsv
= NULL
;
489 updates
= trans
->delayed_ref_updates
;
490 trans
->delayed_ref_updates
= 0;
492 err
= btrfs_run_delayed_refs(trans
, root
, updates
);
493 if (err
) /* Error code will also eval true */
497 trans
->block_rsv
= rsv
;
499 return should_end_transaction(trans
, root
);
502 static int __btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
503 struct btrfs_root
*root
, int throttle
, int lock
)
505 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
506 struct btrfs_fs_info
*info
= root
->fs_info
;
510 if (--trans
->use_count
) {
511 trans
->block_rsv
= trans
->orig_rsv
;
515 btrfs_trans_release_metadata(trans
, root
);
516 trans
->block_rsv
= NULL
;
518 * the same root has to be passed to start_transaction and
519 * end_transaction. Subvolume quota depends on this.
521 WARN_ON(trans
->root
!= root
);
523 unsigned long cur
= trans
->delayed_ref_updates
;
524 trans
->delayed_ref_updates
= 0;
526 trans
->transaction
->delayed_refs
.num_heads_ready
> 64) {
527 trans
->delayed_ref_updates
= 0;
528 btrfs_run_delayed_refs(trans
, root
, cur
);
535 if (lock
&& !atomic_read(&root
->fs_info
->open_ioctl_trans
) &&
536 should_end_transaction(trans
, root
)) {
537 trans
->transaction
->blocked
= 1;
541 if (lock
&& cur_trans
->blocked
&& !cur_trans
->in_commit
) {
544 * We may race with somebody else here so end up having
545 * to call end_transaction on ourselves again, so inc
549 return btrfs_commit_transaction(trans
, root
);
551 wake_up_process(info
->transaction_kthread
);
555 WARN_ON(cur_trans
!= info
->running_transaction
);
556 WARN_ON(atomic_read(&cur_trans
->num_writers
) < 1);
557 atomic_dec(&cur_trans
->num_writers
);
560 if (waitqueue_active(&cur_trans
->writer_wait
))
561 wake_up(&cur_trans
->writer_wait
);
562 put_transaction(cur_trans
);
564 if (current
->journal_info
== trans
)
565 current
->journal_info
= NULL
;
568 btrfs_run_delayed_iputs(root
);
570 if (trans
->aborted
||
571 root
->fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
) {
575 memset(trans
, 0, sizeof(*trans
));
576 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
580 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
581 struct btrfs_root
*root
)
585 ret
= __btrfs_end_transaction(trans
, root
, 0, 1);
591 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
,
592 struct btrfs_root
*root
)
596 ret
= __btrfs_end_transaction(trans
, root
, 1, 1);
602 int btrfs_end_transaction_nolock(struct btrfs_trans_handle
*trans
,
603 struct btrfs_root
*root
)
607 ret
= __btrfs_end_transaction(trans
, root
, 0, 0);
613 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle
*trans
,
614 struct btrfs_root
*root
)
616 return __btrfs_end_transaction(trans
, root
, 1, 1);
620 * when btree blocks are allocated, they have some corresponding bits set for
621 * them in one of two extent_io trees. This is used to make sure all of
622 * those extents are sent to disk but does not wait on them
624 int btrfs_write_marked_extents(struct btrfs_root
*root
,
625 struct extent_io_tree
*dirty_pages
, int mark
)
629 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
633 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
635 convert_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
, mark
,
637 err
= filemap_fdatawrite_range(mapping
, start
, end
);
649 * when btree blocks are allocated, they have some corresponding bits set for
650 * them in one of two extent_io trees. This is used to make sure all of
651 * those extents are on disk for transaction or log commit. We wait
652 * on all the pages and clear them from the dirty pages state tree
654 int btrfs_wait_marked_extents(struct btrfs_root
*root
,
655 struct extent_io_tree
*dirty_pages
, int mark
)
659 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
663 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
665 clear_extent_bits(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
, GFP_NOFS
);
666 err
= filemap_fdatawait_range(mapping
, start
, end
);
678 * when btree blocks are allocated, they have some corresponding bits set for
679 * them in one of two extent_io trees. This is used to make sure all of
680 * those extents are on disk for transaction or log commit
682 int btrfs_write_and_wait_marked_extents(struct btrfs_root
*root
,
683 struct extent_io_tree
*dirty_pages
, int mark
)
688 ret
= btrfs_write_marked_extents(root
, dirty_pages
, mark
);
689 ret2
= btrfs_wait_marked_extents(root
, dirty_pages
, mark
);
698 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
,
699 struct btrfs_root
*root
)
701 if (!trans
|| !trans
->transaction
) {
702 struct inode
*btree_inode
;
703 btree_inode
= root
->fs_info
->btree_inode
;
704 return filemap_write_and_wait(btree_inode
->i_mapping
);
706 return btrfs_write_and_wait_marked_extents(root
,
707 &trans
->transaction
->dirty_pages
,
712 * this is used to update the root pointer in the tree of tree roots.
714 * But, in the case of the extent allocation tree, updating the root
715 * pointer may allocate blocks which may change the root of the extent
718 * So, this loops and repeats and makes sure the cowonly root didn't
719 * change while the root pointer was being updated in the metadata.
721 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
722 struct btrfs_root
*root
)
727 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
729 old_root_used
= btrfs_root_used(&root
->root_item
);
730 btrfs_write_dirty_block_groups(trans
, root
);
733 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
734 if (old_root_bytenr
== root
->node
->start
&&
735 old_root_used
== btrfs_root_used(&root
->root_item
))
738 btrfs_set_root_node(&root
->root_item
, root
->node
);
739 ret
= btrfs_update_root(trans
, tree_root
,
745 old_root_used
= btrfs_root_used(&root
->root_item
);
746 ret
= btrfs_write_dirty_block_groups(trans
, root
);
751 if (root
!= root
->fs_info
->extent_root
)
752 switch_commit_root(root
);
758 * update all the cowonly tree roots on disk
760 * The error handling in this function may not be obvious. Any of the
761 * failures will cause the file system to go offline. We still need
762 * to clean up the delayed refs.
764 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
,
765 struct btrfs_root
*root
)
767 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
768 struct list_head
*next
;
769 struct extent_buffer
*eb
;
772 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
776 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
777 ret
= btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
,
779 btrfs_tree_unlock(eb
);
780 free_extent_buffer(eb
);
785 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
789 ret
= btrfs_run_dev_stats(trans
, root
->fs_info
);
792 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
793 next
= fs_info
->dirty_cowonly_roots
.next
;
795 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
797 ret
= update_cowonly_root(trans
, root
);
802 down_write(&fs_info
->extent_commit_sem
);
803 switch_commit_root(fs_info
->extent_root
);
804 up_write(&fs_info
->extent_commit_sem
);
810 * dead roots are old snapshots that need to be deleted. This allocates
811 * a dirty root struct and adds it into the list of dead roots that need to
814 int btrfs_add_dead_root(struct btrfs_root
*root
)
816 spin_lock(&root
->fs_info
->trans_lock
);
817 list_add(&root
->root_list
, &root
->fs_info
->dead_roots
);
818 spin_unlock(&root
->fs_info
->trans_lock
);
823 * update all the cowonly tree roots on disk
825 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
,
826 struct btrfs_root
*root
)
828 struct btrfs_root
*gang
[8];
829 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
834 spin_lock(&fs_info
->fs_roots_radix_lock
);
836 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
839 BTRFS_ROOT_TRANS_TAG
);
842 for (i
= 0; i
< ret
; i
++) {
844 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
845 (unsigned long)root
->root_key
.objectid
,
846 BTRFS_ROOT_TRANS_TAG
);
847 spin_unlock(&fs_info
->fs_roots_radix_lock
);
849 btrfs_free_log(trans
, root
);
850 btrfs_update_reloc_root(trans
, root
);
851 btrfs_orphan_commit_root(trans
, root
);
853 btrfs_save_ino_cache(root
, trans
);
855 /* see comments in should_cow_block() */
859 if (root
->commit_root
!= root
->node
) {
860 mutex_lock(&root
->fs_commit_mutex
);
861 switch_commit_root(root
);
862 btrfs_unpin_free_ino(root
);
863 mutex_unlock(&root
->fs_commit_mutex
);
865 btrfs_set_root_node(&root
->root_item
,
869 err
= btrfs_update_root(trans
, fs_info
->tree_root
,
872 spin_lock(&fs_info
->fs_roots_radix_lock
);
877 spin_unlock(&fs_info
->fs_roots_radix_lock
);
882 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
883 * otherwise every leaf in the btree is read and defragged.
885 int btrfs_defrag_root(struct btrfs_root
*root
, int cacheonly
)
887 struct btrfs_fs_info
*info
= root
->fs_info
;
888 struct btrfs_trans_handle
*trans
;
892 if (xchg(&root
->defrag_running
, 1))
896 trans
= btrfs_start_transaction(root
, 0);
898 return PTR_ERR(trans
);
900 ret
= btrfs_defrag_leaves(trans
, root
, cacheonly
);
902 nr
= trans
->blocks_used
;
903 btrfs_end_transaction(trans
, root
);
904 btrfs_btree_balance_dirty(info
->tree_root
, nr
);
907 if (btrfs_fs_closing(root
->fs_info
) || ret
!= -EAGAIN
)
910 root
->defrag_running
= 0;
915 * new snapshots need to be created at a very specific time in the
916 * transaction commit. This does the actual creation
918 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
919 struct btrfs_fs_info
*fs_info
,
920 struct btrfs_pending_snapshot
*pending
)
922 struct btrfs_key key
;
923 struct btrfs_root_item
*new_root_item
;
924 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
925 struct btrfs_root
*root
= pending
->root
;
926 struct btrfs_root
*parent_root
;
927 struct btrfs_block_rsv
*rsv
;
928 struct inode
*parent_inode
;
929 struct dentry
*parent
;
930 struct dentry
*dentry
;
931 struct extent_buffer
*tmp
;
932 struct extent_buffer
*old
;
939 rsv
= trans
->block_rsv
;
941 new_root_item
= kmalloc(sizeof(*new_root_item
), GFP_NOFS
);
942 if (!new_root_item
) {
943 ret
= pending
->error
= -ENOMEM
;
947 ret
= btrfs_find_free_objectid(tree_root
, &objectid
);
949 pending
->error
= ret
;
953 btrfs_reloc_pre_snapshot(trans
, pending
, &to_reserve
);
955 if (to_reserve
> 0) {
956 ret
= btrfs_block_rsv_add_noflush(root
, &pending
->block_rsv
,
959 pending
->error
= ret
;
964 key
.objectid
= objectid
;
965 key
.offset
= (u64
)-1;
966 key
.type
= BTRFS_ROOT_ITEM_KEY
;
968 trans
->block_rsv
= &pending
->block_rsv
;
970 dentry
= pending
->dentry
;
971 parent
= dget_parent(dentry
);
972 parent_inode
= parent
->d_inode
;
973 parent_root
= BTRFS_I(parent_inode
)->root
;
974 record_root_in_trans(trans
, parent_root
);
977 * insert the directory item
979 ret
= btrfs_set_inode_index(parent_inode
, &index
);
980 BUG_ON(ret
); /* -ENOMEM */
981 ret
= btrfs_insert_dir_item(trans
, parent_root
,
982 dentry
->d_name
.name
, dentry
->d_name
.len
,
984 BTRFS_FT_DIR
, index
);
985 if (ret
== -EEXIST
) {
986 pending
->error
= -EEXIST
;
990 goto abort_trans_dput
;
993 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
994 dentry
->d_name
.len
* 2);
995 ret
= btrfs_update_inode(trans
, parent_root
, parent_inode
);
997 goto abort_trans_dput
;
1000 * pull in the delayed directory update
1001 * and the delayed inode item
1002 * otherwise we corrupt the FS during
1005 ret
= btrfs_run_delayed_items(trans
, root
);
1006 if (ret
) { /* Transaction aborted */
1011 record_root_in_trans(trans
, root
);
1012 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
1013 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
1014 btrfs_check_and_init_root_item(new_root_item
);
1016 root_flags
= btrfs_root_flags(new_root_item
);
1017 if (pending
->readonly
)
1018 root_flags
|= BTRFS_ROOT_SUBVOL_RDONLY
;
1020 root_flags
&= ~BTRFS_ROOT_SUBVOL_RDONLY
;
1021 btrfs_set_root_flags(new_root_item
, root_flags
);
1023 old
= btrfs_lock_root_node(root
);
1024 ret
= btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
);
1026 btrfs_tree_unlock(old
);
1027 free_extent_buffer(old
);
1028 goto abort_trans_dput
;
1031 btrfs_set_lock_blocking(old
);
1033 ret
= btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
1034 /* clean up in any case */
1035 btrfs_tree_unlock(old
);
1036 free_extent_buffer(old
);
1038 goto abort_trans_dput
;
1040 /* see comments in should_cow_block() */
1041 root
->force_cow
= 1;
1044 btrfs_set_root_node(new_root_item
, tmp
);
1045 /* record when the snapshot was created in key.offset */
1046 key
.offset
= trans
->transid
;
1047 ret
= btrfs_insert_root(trans
, tree_root
, &key
, new_root_item
);
1048 btrfs_tree_unlock(tmp
);
1049 free_extent_buffer(tmp
);
1051 goto abort_trans_dput
;
1054 * insert root back/forward references
1056 ret
= btrfs_add_root_ref(trans
, tree_root
, objectid
,
1057 parent_root
->root_key
.objectid
,
1058 btrfs_ino(parent_inode
), index
,
1059 dentry
->d_name
.name
, dentry
->d_name
.len
);
1064 key
.offset
= (u64
)-1;
1065 pending
->snap
= btrfs_read_fs_root_no_name(root
->fs_info
, &key
);
1066 if (IS_ERR(pending
->snap
)) {
1067 ret
= PTR_ERR(pending
->snap
);
1071 ret
= btrfs_reloc_post_snapshot(trans
, pending
);
1076 kfree(new_root_item
);
1077 trans
->block_rsv
= rsv
;
1078 btrfs_block_rsv_release(root
, &pending
->block_rsv
, (u64
)-1);
1084 btrfs_abort_transaction(trans
, root
, ret
);
1089 * create all the snapshots we've scheduled for creation
1091 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
,
1092 struct btrfs_fs_info
*fs_info
)
1094 struct btrfs_pending_snapshot
*pending
;
1095 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
1097 list_for_each_entry(pending
, head
, list
)
1098 create_pending_snapshot(trans
, fs_info
, pending
);
1102 static void update_super_roots(struct btrfs_root
*root
)
1104 struct btrfs_root_item
*root_item
;
1105 struct btrfs_super_block
*super
;
1107 super
= root
->fs_info
->super_copy
;
1109 root_item
= &root
->fs_info
->chunk_root
->root_item
;
1110 super
->chunk_root
= root_item
->bytenr
;
1111 super
->chunk_root_generation
= root_item
->generation
;
1112 super
->chunk_root_level
= root_item
->level
;
1114 root_item
= &root
->fs_info
->tree_root
->root_item
;
1115 super
->root
= root_item
->bytenr
;
1116 super
->generation
= root_item
->generation
;
1117 super
->root_level
= root_item
->level
;
1118 if (btrfs_test_opt(root
, SPACE_CACHE
))
1119 super
->cache_generation
= root_item
->generation
;
1122 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
1125 spin_lock(&info
->trans_lock
);
1126 if (info
->running_transaction
)
1127 ret
= info
->running_transaction
->in_commit
;
1128 spin_unlock(&info
->trans_lock
);
1132 int btrfs_transaction_blocked(struct btrfs_fs_info
*info
)
1135 spin_lock(&info
->trans_lock
);
1136 if (info
->running_transaction
)
1137 ret
= info
->running_transaction
->blocked
;
1138 spin_unlock(&info
->trans_lock
);
1143 * wait for the current transaction commit to start and block subsequent
1146 static void wait_current_trans_commit_start(struct btrfs_root
*root
,
1147 struct btrfs_transaction
*trans
)
1149 wait_event(root
->fs_info
->transaction_blocked_wait
, trans
->in_commit
);
1153 * wait for the current transaction to start and then become unblocked.
1156 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root
*root
,
1157 struct btrfs_transaction
*trans
)
1159 wait_event(root
->fs_info
->transaction_wait
,
1160 trans
->commit_done
|| (trans
->in_commit
&& !trans
->blocked
));
1164 * commit transactions asynchronously. once btrfs_commit_transaction_async
1165 * returns, any subsequent transaction will not be allowed to join.
1167 struct btrfs_async_commit
{
1168 struct btrfs_trans_handle
*newtrans
;
1169 struct btrfs_root
*root
;
1170 struct delayed_work work
;
1173 static void do_async_commit(struct work_struct
*work
)
1175 struct btrfs_async_commit
*ac
=
1176 container_of(work
, struct btrfs_async_commit
, work
.work
);
1178 btrfs_commit_transaction(ac
->newtrans
, ac
->root
);
1182 int btrfs_commit_transaction_async(struct btrfs_trans_handle
*trans
,
1183 struct btrfs_root
*root
,
1184 int wait_for_unblock
)
1186 struct btrfs_async_commit
*ac
;
1187 struct btrfs_transaction
*cur_trans
;
1189 ac
= kmalloc(sizeof(*ac
), GFP_NOFS
);
1193 INIT_DELAYED_WORK(&ac
->work
, do_async_commit
);
1195 ac
->newtrans
= btrfs_join_transaction(root
);
1196 if (IS_ERR(ac
->newtrans
)) {
1197 int err
= PTR_ERR(ac
->newtrans
);
1202 /* take transaction reference */
1203 cur_trans
= trans
->transaction
;
1204 atomic_inc(&cur_trans
->use_count
);
1206 btrfs_end_transaction(trans
, root
);
1207 schedule_delayed_work(&ac
->work
, 0);
1209 /* wait for transaction to start and unblock */
1210 if (wait_for_unblock
)
1211 wait_current_trans_commit_start_and_unblock(root
, cur_trans
);
1213 wait_current_trans_commit_start(root
, cur_trans
);
1215 if (current
->journal_info
== trans
)
1216 current
->journal_info
= NULL
;
1218 put_transaction(cur_trans
);
1223 static void cleanup_transaction(struct btrfs_trans_handle
*trans
,
1224 struct btrfs_root
*root
, int err
)
1226 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1228 WARN_ON(trans
->use_count
> 1);
1230 btrfs_abort_transaction(trans
, root
, err
);
1232 spin_lock(&root
->fs_info
->trans_lock
);
1233 list_del_init(&cur_trans
->list
);
1234 if (cur_trans
== root
->fs_info
->running_transaction
) {
1235 root
->fs_info
->running_transaction
= NULL
;
1236 root
->fs_info
->trans_no_join
= 0;
1238 spin_unlock(&root
->fs_info
->trans_lock
);
1240 btrfs_cleanup_one_transaction(trans
->transaction
, root
);
1242 put_transaction(cur_trans
);
1243 put_transaction(cur_trans
);
1245 trace_btrfs_transaction_commit(root
);
1247 btrfs_scrub_continue(root
);
1249 if (current
->journal_info
== trans
)
1250 current
->journal_info
= NULL
;
1252 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1256 * btrfs_transaction state sequence:
1257 * in_commit = 0, blocked = 0 (initial)
1258 * in_commit = 1, blocked = 1
1262 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
,
1263 struct btrfs_root
*root
)
1265 unsigned long joined
= 0;
1266 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1267 struct btrfs_transaction
*prev_trans
= NULL
;
1270 int should_grow
= 0;
1271 unsigned long now
= get_seconds();
1272 int flush_on_commit
= btrfs_test_opt(root
, FLUSHONCOMMIT
);
1274 btrfs_run_ordered_operations(root
, 0);
1276 btrfs_trans_release_metadata(trans
, root
);
1277 trans
->block_rsv
= NULL
;
1279 if (cur_trans
->aborted
)
1280 goto cleanup_transaction
;
1282 /* make a pass through all the delayed refs we have so far
1283 * any runnings procs may add more while we are here
1285 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1287 goto cleanup_transaction
;
1289 cur_trans
= trans
->transaction
;
1292 * set the flushing flag so procs in this transaction have to
1293 * start sending their work down.
1295 cur_trans
->delayed_refs
.flushing
= 1;
1297 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1299 goto cleanup_transaction
;
1301 spin_lock(&cur_trans
->commit_lock
);
1302 if (cur_trans
->in_commit
) {
1303 spin_unlock(&cur_trans
->commit_lock
);
1304 atomic_inc(&cur_trans
->use_count
);
1305 ret
= btrfs_end_transaction(trans
, root
);
1307 wait_for_commit(root
, cur_trans
);
1309 put_transaction(cur_trans
);
1314 trans
->transaction
->in_commit
= 1;
1315 trans
->transaction
->blocked
= 1;
1316 spin_unlock(&cur_trans
->commit_lock
);
1317 wake_up(&root
->fs_info
->transaction_blocked_wait
);
1319 spin_lock(&root
->fs_info
->trans_lock
);
1320 if (cur_trans
->list
.prev
!= &root
->fs_info
->trans_list
) {
1321 prev_trans
= list_entry(cur_trans
->list
.prev
,
1322 struct btrfs_transaction
, list
);
1323 if (!prev_trans
->commit_done
) {
1324 atomic_inc(&prev_trans
->use_count
);
1325 spin_unlock(&root
->fs_info
->trans_lock
);
1327 wait_for_commit(root
, prev_trans
);
1329 put_transaction(prev_trans
);
1331 spin_unlock(&root
->fs_info
->trans_lock
);
1334 spin_unlock(&root
->fs_info
->trans_lock
);
1337 if (now
< cur_trans
->start_time
|| now
- cur_trans
->start_time
< 1)
1341 int snap_pending
= 0;
1343 joined
= cur_trans
->num_joined
;
1344 if (!list_empty(&trans
->transaction
->pending_snapshots
))
1347 WARN_ON(cur_trans
!= trans
->transaction
);
1349 if (flush_on_commit
|| snap_pending
) {
1350 btrfs_start_delalloc_inodes(root
, 1);
1351 btrfs_wait_ordered_extents(root
, 0, 1);
1354 ret
= btrfs_run_delayed_items(trans
, root
);
1356 goto cleanup_transaction
;
1359 * rename don't use btrfs_join_transaction, so, once we
1360 * set the transaction to blocked above, we aren't going
1361 * to get any new ordered operations. We can safely run
1362 * it here and no for sure that nothing new will be added
1365 btrfs_run_ordered_operations(root
, 1);
1367 prepare_to_wait(&cur_trans
->writer_wait
, &wait
,
1368 TASK_UNINTERRUPTIBLE
);
1370 if (atomic_read(&cur_trans
->num_writers
) > 1)
1371 schedule_timeout(MAX_SCHEDULE_TIMEOUT
);
1372 else if (should_grow
)
1373 schedule_timeout(1);
1375 finish_wait(&cur_trans
->writer_wait
, &wait
);
1376 } while (atomic_read(&cur_trans
->num_writers
) > 1 ||
1377 (should_grow
&& cur_trans
->num_joined
!= joined
));
1380 * Ok now we need to make sure to block out any other joins while we
1381 * commit the transaction. We could have started a join before setting
1382 * no_join so make sure to wait for num_writers to == 1 again.
1384 spin_lock(&root
->fs_info
->trans_lock
);
1385 root
->fs_info
->trans_no_join
= 1;
1386 spin_unlock(&root
->fs_info
->trans_lock
);
1387 wait_event(cur_trans
->writer_wait
,
1388 atomic_read(&cur_trans
->num_writers
) == 1);
1391 * the reloc mutex makes sure that we stop
1392 * the balancing code from coming in and moving
1393 * extents around in the middle of the commit
1395 mutex_lock(&root
->fs_info
->reloc_mutex
);
1397 ret
= btrfs_run_delayed_items(trans
, root
);
1399 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1400 goto cleanup_transaction
;
1403 ret
= create_pending_snapshots(trans
, root
->fs_info
);
1405 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1406 goto cleanup_transaction
;
1409 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1411 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1412 goto cleanup_transaction
;
1416 * make sure none of the code above managed to slip in a
1419 btrfs_assert_delayed_root_empty(root
);
1421 WARN_ON(cur_trans
!= trans
->transaction
);
1423 btrfs_scrub_pause(root
);
1424 /* btrfs_commit_tree_roots is responsible for getting the
1425 * various roots consistent with each other. Every pointer
1426 * in the tree of tree roots has to point to the most up to date
1427 * root for every subvolume and other tree. So, we have to keep
1428 * the tree logging code from jumping in and changing any
1431 * At this point in the commit, there can't be any tree-log
1432 * writers, but a little lower down we drop the trans mutex
1433 * and let new people in. By holding the tree_log_mutex
1434 * from now until after the super is written, we avoid races
1435 * with the tree-log code.
1437 mutex_lock(&root
->fs_info
->tree_log_mutex
);
1439 ret
= commit_fs_roots(trans
, root
);
1441 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1442 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1443 goto cleanup_transaction
;
1446 /* commit_fs_roots gets rid of all the tree log roots, it is now
1447 * safe to free the root of tree log roots
1449 btrfs_free_log_root_tree(trans
, root
->fs_info
);
1451 ret
= commit_cowonly_roots(trans
, root
);
1453 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1454 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1455 goto cleanup_transaction
;
1458 btrfs_prepare_extent_commit(trans
, root
);
1460 cur_trans
= root
->fs_info
->running_transaction
;
1462 btrfs_set_root_node(&root
->fs_info
->tree_root
->root_item
,
1463 root
->fs_info
->tree_root
->node
);
1464 switch_commit_root(root
->fs_info
->tree_root
);
1466 btrfs_set_root_node(&root
->fs_info
->chunk_root
->root_item
,
1467 root
->fs_info
->chunk_root
->node
);
1468 switch_commit_root(root
->fs_info
->chunk_root
);
1470 update_super_roots(root
);
1472 if (!root
->fs_info
->log_root_recovering
) {
1473 btrfs_set_super_log_root(root
->fs_info
->super_copy
, 0);
1474 btrfs_set_super_log_root_level(root
->fs_info
->super_copy
, 0);
1477 memcpy(root
->fs_info
->super_for_commit
, root
->fs_info
->super_copy
,
1478 sizeof(*root
->fs_info
->super_copy
));
1480 trans
->transaction
->blocked
= 0;
1481 spin_lock(&root
->fs_info
->trans_lock
);
1482 root
->fs_info
->running_transaction
= NULL
;
1483 root
->fs_info
->trans_no_join
= 0;
1484 spin_unlock(&root
->fs_info
->trans_lock
);
1485 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1487 wake_up(&root
->fs_info
->transaction_wait
);
1489 ret
= btrfs_write_and_wait_transaction(trans
, root
);
1491 btrfs_error(root
->fs_info
, ret
,
1492 "Error while writing out transaction.");
1493 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1494 goto cleanup_transaction
;
1497 ret
= write_ctree_super(trans
, root
, 0);
1499 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1500 goto cleanup_transaction
;
1504 * the super is written, we can safely allow the tree-loggers
1505 * to go about their business
1507 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1509 btrfs_finish_extent_commit(trans
, root
);
1511 cur_trans
->commit_done
= 1;
1513 root
->fs_info
->last_trans_committed
= cur_trans
->transid
;
1515 wake_up(&cur_trans
->commit_wait
);
1517 spin_lock(&root
->fs_info
->trans_lock
);
1518 list_del_init(&cur_trans
->list
);
1519 spin_unlock(&root
->fs_info
->trans_lock
);
1521 put_transaction(cur_trans
);
1522 put_transaction(cur_trans
);
1524 trace_btrfs_transaction_commit(root
);
1526 btrfs_scrub_continue(root
);
1528 if (current
->journal_info
== trans
)
1529 current
->journal_info
= NULL
;
1531 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1533 if (current
!= root
->fs_info
->transaction_kthread
)
1534 btrfs_run_delayed_iputs(root
);
1538 cleanup_transaction
:
1539 btrfs_printk(root
->fs_info
, "Skipping commit of aborted transaction.\n");
1541 if (current
->journal_info
== trans
)
1542 current
->journal_info
= NULL
;
1543 cleanup_transaction(trans
, root
, ret
);
1549 * interface function to delete all the snapshots we have scheduled for deletion
1551 int btrfs_clean_old_snapshots(struct btrfs_root
*root
)
1554 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1556 spin_lock(&fs_info
->trans_lock
);
1557 list_splice_init(&fs_info
->dead_roots
, &list
);
1558 spin_unlock(&fs_info
->trans_lock
);
1560 while (!list_empty(&list
)) {
1563 root
= list_entry(list
.next
, struct btrfs_root
, root_list
);
1564 list_del(&root
->root_list
);
1566 btrfs_kill_all_delayed_nodes(root
);
1568 if (btrfs_header_backref_rev(root
->node
) <
1569 BTRFS_MIXED_BACKREF_REV
)
1570 ret
= btrfs_drop_snapshot(root
, NULL
, 0, 0);
1572 ret
=btrfs_drop_snapshot(root
, NULL
, 1, 0);