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 WARN_ON(!list_empty(&transaction
->delayed_refs
.seq_head
));
42 memset(transaction
, 0, sizeof(*transaction
));
43 kmem_cache_free(btrfs_transaction_cachep
, transaction
);
47 static noinline
void switch_commit_root(struct btrfs_root
*root
)
49 free_extent_buffer(root
->commit_root
);
50 root
->commit_root
= btrfs_root_node(root
);
54 * either allocate a new transaction or hop into the existing one
56 static noinline
int join_transaction(struct btrfs_root
*root
, int nofail
)
58 struct btrfs_transaction
*cur_trans
;
60 spin_lock(&root
->fs_info
->trans_lock
);
62 /* The file system has been taken offline. No new transactions. */
63 if (root
->fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
) {
64 spin_unlock(&root
->fs_info
->trans_lock
);
68 if (root
->fs_info
->trans_no_join
) {
70 spin_unlock(&root
->fs_info
->trans_lock
);
75 cur_trans
= root
->fs_info
->running_transaction
;
77 if (cur_trans
->aborted
) {
78 spin_unlock(&root
->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(&root
->fs_info
->trans_lock
);
87 spin_unlock(&root
->fs_info
->trans_lock
);
89 cur_trans
= kmem_cache_alloc(btrfs_transaction_cachep
, GFP_NOFS
);
93 spin_lock(&root
->fs_info
->trans_lock
);
94 if (root
->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
= root
->fs_info
->running_transaction
;
104 atomic_set(&cur_trans
->num_writers
, 1);
105 cur_trans
->num_joined
= 0;
106 init_waitqueue_head(&cur_trans
->writer_wait
);
107 init_waitqueue_head(&cur_trans
->commit_wait
);
108 cur_trans
->in_commit
= 0;
109 cur_trans
->blocked
= 0;
111 * One for this trans handle, one so it will live on until we
112 * commit the transaction.
114 atomic_set(&cur_trans
->use_count
, 2);
115 cur_trans
->commit_done
= 0;
116 cur_trans
->start_time
= get_seconds();
118 cur_trans
->delayed_refs
.root
= RB_ROOT
;
119 cur_trans
->delayed_refs
.num_entries
= 0;
120 cur_trans
->delayed_refs
.num_heads_ready
= 0;
121 cur_trans
->delayed_refs
.num_heads
= 0;
122 cur_trans
->delayed_refs
.flushing
= 0;
123 cur_trans
->delayed_refs
.run_delayed_start
= 0;
124 cur_trans
->delayed_refs
.seq
= 1;
125 init_waitqueue_head(&cur_trans
->delayed_refs
.seq_wait
);
126 spin_lock_init(&cur_trans
->commit_lock
);
127 spin_lock_init(&cur_trans
->delayed_refs
.lock
);
128 INIT_LIST_HEAD(&cur_trans
->delayed_refs
.seq_head
);
130 INIT_LIST_HEAD(&cur_trans
->pending_snapshots
);
131 list_add_tail(&cur_trans
->list
, &root
->fs_info
->trans_list
);
132 extent_io_tree_init(&cur_trans
->dirty_pages
,
133 root
->fs_info
->btree_inode
->i_mapping
);
134 root
->fs_info
->generation
++;
135 cur_trans
->transid
= root
->fs_info
->generation
;
136 root
->fs_info
->running_transaction
= cur_trans
;
137 cur_trans
->aborted
= 0;
138 spin_unlock(&root
->fs_info
->trans_lock
);
144 * this does all the record keeping required to make sure that a reference
145 * counted root is properly recorded in a given transaction. This is required
146 * to make sure the old root from before we joined the transaction is deleted
147 * when the transaction commits
149 static int record_root_in_trans(struct btrfs_trans_handle
*trans
,
150 struct btrfs_root
*root
)
152 if (root
->ref_cows
&& root
->last_trans
< trans
->transid
) {
153 WARN_ON(root
== root
->fs_info
->extent_root
);
154 WARN_ON(root
->commit_root
!= root
->node
);
157 * see below for in_trans_setup usage rules
158 * we have the reloc mutex held now, so there
159 * is only one writer in this function
161 root
->in_trans_setup
= 1;
163 /* make sure readers find in_trans_setup before
164 * they find our root->last_trans update
168 spin_lock(&root
->fs_info
->fs_roots_radix_lock
);
169 if (root
->last_trans
== trans
->transid
) {
170 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
173 radix_tree_tag_set(&root
->fs_info
->fs_roots_radix
,
174 (unsigned long)root
->root_key
.objectid
,
175 BTRFS_ROOT_TRANS_TAG
);
176 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
177 root
->last_trans
= trans
->transid
;
179 /* this is pretty tricky. We don't want to
180 * take the relocation lock in btrfs_record_root_in_trans
181 * unless we're really doing the first setup for this root in
184 * Normally we'd use root->last_trans as a flag to decide
185 * if we want to take the expensive mutex.
187 * But, we have to set root->last_trans before we
188 * init the relocation root, otherwise, we trip over warnings
189 * in ctree.c. The solution used here is to flag ourselves
190 * with root->in_trans_setup. When this is 1, we're still
191 * fixing up the reloc trees and everyone must wait.
193 * When this is zero, they can trust root->last_trans and fly
194 * through btrfs_record_root_in_trans without having to take the
195 * lock. smp_wmb() makes sure that all the writes above are
196 * done before we pop in the zero below
198 btrfs_init_reloc_root(trans
, root
);
200 root
->in_trans_setup
= 0;
206 int btrfs_record_root_in_trans(struct btrfs_trans_handle
*trans
,
207 struct btrfs_root
*root
)
213 * see record_root_in_trans for comments about in_trans_setup usage
217 if (root
->last_trans
== trans
->transid
&&
218 !root
->in_trans_setup
)
221 mutex_lock(&root
->fs_info
->reloc_mutex
);
222 record_root_in_trans(trans
, root
);
223 mutex_unlock(&root
->fs_info
->reloc_mutex
);
228 /* wait for commit against the current transaction to become unblocked
229 * when this is done, it is safe to start a new transaction, but the current
230 * transaction might not be fully on disk.
232 static void wait_current_trans(struct btrfs_root
*root
)
234 struct btrfs_transaction
*cur_trans
;
236 spin_lock(&root
->fs_info
->trans_lock
);
237 cur_trans
= root
->fs_info
->running_transaction
;
238 if (cur_trans
&& cur_trans
->blocked
) {
239 atomic_inc(&cur_trans
->use_count
);
240 spin_unlock(&root
->fs_info
->trans_lock
);
242 wait_event(root
->fs_info
->transaction_wait
,
243 !cur_trans
->blocked
);
244 put_transaction(cur_trans
);
246 spin_unlock(&root
->fs_info
->trans_lock
);
250 enum btrfs_trans_type
{
257 static int may_wait_transaction(struct btrfs_root
*root
, int type
)
259 if (root
->fs_info
->log_root_recovering
)
262 if (type
== TRANS_USERSPACE
)
265 if (type
== TRANS_START
&&
266 !atomic_read(&root
->fs_info
->open_ioctl_trans
))
272 static struct btrfs_trans_handle
*start_transaction(struct btrfs_root
*root
,
273 u64 num_items
, int type
)
275 struct btrfs_trans_handle
*h
;
276 struct btrfs_transaction
*cur_trans
;
280 if (root
->fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
)
281 return ERR_PTR(-EROFS
);
283 if (current
->journal_info
) {
284 WARN_ON(type
!= TRANS_JOIN
&& type
!= TRANS_JOIN_NOLOCK
);
285 h
= current
->journal_info
;
287 h
->orig_rsv
= h
->block_rsv
;
293 * Do the reservation before we join the transaction so we can do all
294 * the appropriate flushing if need be.
296 if (num_items
> 0 && root
!= root
->fs_info
->chunk_root
) {
297 num_bytes
= btrfs_calc_trans_metadata_size(root
, num_items
);
298 ret
= btrfs_block_rsv_add(root
,
299 &root
->fs_info
->trans_block_rsv
,
305 h
= kmem_cache_alloc(btrfs_trans_handle_cachep
, GFP_NOFS
);
307 return ERR_PTR(-ENOMEM
);
309 if (may_wait_transaction(root
, type
))
310 wait_current_trans(root
);
313 ret
= join_transaction(root
, type
== TRANS_JOIN_NOLOCK
);
315 wait_current_trans(root
);
316 } while (ret
== -EBUSY
);
319 kmem_cache_free(btrfs_trans_handle_cachep
, h
);
323 cur_trans
= root
->fs_info
->running_transaction
;
325 h
->transid
= cur_trans
->transid
;
326 h
->transaction
= cur_trans
;
328 h
->bytes_reserved
= 0;
329 h
->delayed_ref_updates
= 0;
336 if (cur_trans
->blocked
&& may_wait_transaction(root
, type
)) {
337 btrfs_commit_transaction(h
, root
);
342 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
343 h
->transid
, num_bytes
, 1);
344 h
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
345 h
->bytes_reserved
= num_bytes
;
349 btrfs_record_root_in_trans(h
, root
);
351 if (!current
->journal_info
&& type
!= TRANS_USERSPACE
)
352 current
->journal_info
= h
;
356 struct btrfs_trans_handle
*btrfs_start_transaction(struct btrfs_root
*root
,
359 return start_transaction(root
, num_items
, TRANS_START
);
361 struct btrfs_trans_handle
*btrfs_join_transaction(struct btrfs_root
*root
)
363 return start_transaction(root
, 0, TRANS_JOIN
);
366 struct btrfs_trans_handle
*btrfs_join_transaction_nolock(struct btrfs_root
*root
)
368 return start_transaction(root
, 0, TRANS_JOIN_NOLOCK
);
371 struct btrfs_trans_handle
*btrfs_start_ioctl_transaction(struct btrfs_root
*root
)
373 return start_transaction(root
, 0, TRANS_USERSPACE
);
376 /* wait for a transaction commit to be fully complete */
377 static noinline
void wait_for_commit(struct btrfs_root
*root
,
378 struct btrfs_transaction
*commit
)
380 wait_event(commit
->commit_wait
, commit
->commit_done
);
383 int btrfs_wait_for_commit(struct btrfs_root
*root
, u64 transid
)
385 struct btrfs_transaction
*cur_trans
= NULL
, *t
;
390 if (transid
<= root
->fs_info
->last_trans_committed
)
393 /* find specified transaction */
394 spin_lock(&root
->fs_info
->trans_lock
);
395 list_for_each_entry(t
, &root
->fs_info
->trans_list
, list
) {
396 if (t
->transid
== transid
) {
398 atomic_inc(&cur_trans
->use_count
);
401 if (t
->transid
> transid
)
404 spin_unlock(&root
->fs_info
->trans_lock
);
407 goto out
; /* bad transid */
409 /* find newest transaction that is committing | committed */
410 spin_lock(&root
->fs_info
->trans_lock
);
411 list_for_each_entry_reverse(t
, &root
->fs_info
->trans_list
,
417 atomic_inc(&cur_trans
->use_count
);
421 spin_unlock(&root
->fs_info
->trans_lock
);
423 goto out
; /* nothing committing|committed */
426 wait_for_commit(root
, cur_trans
);
428 put_transaction(cur_trans
);
434 void btrfs_throttle(struct btrfs_root
*root
)
436 if (!atomic_read(&root
->fs_info
->open_ioctl_trans
))
437 wait_current_trans(root
);
440 static int should_end_transaction(struct btrfs_trans_handle
*trans
,
441 struct btrfs_root
*root
)
445 ret
= btrfs_block_rsv_check(root
, &root
->fs_info
->global_block_rsv
, 5);
449 int btrfs_should_end_transaction(struct btrfs_trans_handle
*trans
,
450 struct btrfs_root
*root
)
452 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
453 struct btrfs_block_rsv
*rsv
= trans
->block_rsv
;
458 if (cur_trans
->blocked
|| cur_trans
->delayed_refs
.flushing
)
462 * We need to do this in case we're deleting csums so the global block
463 * rsv get's used instead of the csum block rsv.
465 trans
->block_rsv
= NULL
;
467 updates
= trans
->delayed_ref_updates
;
468 trans
->delayed_ref_updates
= 0;
470 err
= btrfs_run_delayed_refs(trans
, root
, updates
);
471 if (err
) /* Error code will also eval true */
475 trans
->block_rsv
= rsv
;
477 return should_end_transaction(trans
, root
);
480 static int __btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
481 struct btrfs_root
*root
, int throttle
, int lock
)
483 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
484 struct btrfs_fs_info
*info
= root
->fs_info
;
488 if (--trans
->use_count
) {
489 trans
->block_rsv
= trans
->orig_rsv
;
493 btrfs_trans_release_metadata(trans
, root
);
494 trans
->block_rsv
= NULL
;
496 unsigned long cur
= trans
->delayed_ref_updates
;
497 trans
->delayed_ref_updates
= 0;
499 trans
->transaction
->delayed_refs
.num_heads_ready
> 64) {
500 trans
->delayed_ref_updates
= 0;
501 btrfs_run_delayed_refs(trans
, root
, cur
);
508 if (lock
&& !atomic_read(&root
->fs_info
->open_ioctl_trans
) &&
509 should_end_transaction(trans
, root
)) {
510 trans
->transaction
->blocked
= 1;
514 if (lock
&& cur_trans
->blocked
&& !cur_trans
->in_commit
) {
517 * We may race with somebody else here so end up having
518 * to call end_transaction on ourselves again, so inc
522 return btrfs_commit_transaction(trans
, root
);
524 wake_up_process(info
->transaction_kthread
);
528 WARN_ON(cur_trans
!= info
->running_transaction
);
529 WARN_ON(atomic_read(&cur_trans
->num_writers
) < 1);
530 atomic_dec(&cur_trans
->num_writers
);
533 if (waitqueue_active(&cur_trans
->writer_wait
))
534 wake_up(&cur_trans
->writer_wait
);
535 put_transaction(cur_trans
);
537 if (current
->journal_info
== trans
)
538 current
->journal_info
= NULL
;
541 btrfs_run_delayed_iputs(root
);
543 if (trans
->aborted
||
544 root
->fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
) {
548 memset(trans
, 0, sizeof(*trans
));
549 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
553 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
554 struct btrfs_root
*root
)
558 ret
= __btrfs_end_transaction(trans
, root
, 0, 1);
564 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
,
565 struct btrfs_root
*root
)
569 ret
= __btrfs_end_transaction(trans
, root
, 1, 1);
575 int btrfs_end_transaction_nolock(struct btrfs_trans_handle
*trans
,
576 struct btrfs_root
*root
)
580 ret
= __btrfs_end_transaction(trans
, root
, 0, 0);
586 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle
*trans
,
587 struct btrfs_root
*root
)
589 return __btrfs_end_transaction(trans
, root
, 1, 1);
593 * when btree blocks are allocated, they have some corresponding bits set for
594 * them in one of two extent_io trees. This is used to make sure all of
595 * those extents are sent to disk but does not wait on them
597 int btrfs_write_marked_extents(struct btrfs_root
*root
,
598 struct extent_io_tree
*dirty_pages
, int mark
)
602 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
606 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
608 convert_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
, mark
,
610 err
= filemap_fdatawrite_range(mapping
, start
, end
);
622 * when btree blocks are allocated, they have some corresponding bits set for
623 * them in one of two extent_io trees. This is used to make sure all of
624 * those extents are on disk for transaction or log commit. We wait
625 * on all the pages and clear them from the dirty pages state tree
627 int btrfs_wait_marked_extents(struct btrfs_root
*root
,
628 struct extent_io_tree
*dirty_pages
, int mark
)
632 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
636 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
638 clear_extent_bits(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
, GFP_NOFS
);
639 err
= filemap_fdatawait_range(mapping
, start
, end
);
651 * when btree blocks are allocated, they have some corresponding bits set for
652 * them in one of two extent_io trees. This is used to make sure all of
653 * those extents are on disk for transaction or log commit
655 int btrfs_write_and_wait_marked_extents(struct btrfs_root
*root
,
656 struct extent_io_tree
*dirty_pages
, int mark
)
661 ret
= btrfs_write_marked_extents(root
, dirty_pages
, mark
);
662 ret2
= btrfs_wait_marked_extents(root
, dirty_pages
, mark
);
671 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
,
672 struct btrfs_root
*root
)
674 if (!trans
|| !trans
->transaction
) {
675 struct inode
*btree_inode
;
676 btree_inode
= root
->fs_info
->btree_inode
;
677 return filemap_write_and_wait(btree_inode
->i_mapping
);
679 return btrfs_write_and_wait_marked_extents(root
,
680 &trans
->transaction
->dirty_pages
,
685 * this is used to update the root pointer in the tree of tree roots.
687 * But, in the case of the extent allocation tree, updating the root
688 * pointer may allocate blocks which may change the root of the extent
691 * So, this loops and repeats and makes sure the cowonly root didn't
692 * change while the root pointer was being updated in the metadata.
694 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
695 struct btrfs_root
*root
)
700 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
702 old_root_used
= btrfs_root_used(&root
->root_item
);
703 btrfs_write_dirty_block_groups(trans
, root
);
706 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
707 if (old_root_bytenr
== root
->node
->start
&&
708 old_root_used
== btrfs_root_used(&root
->root_item
))
711 btrfs_set_root_node(&root
->root_item
, root
->node
);
712 ret
= btrfs_update_root(trans
, tree_root
,
718 old_root_used
= btrfs_root_used(&root
->root_item
);
719 ret
= btrfs_write_dirty_block_groups(trans
, root
);
724 if (root
!= root
->fs_info
->extent_root
)
725 switch_commit_root(root
);
731 * update all the cowonly tree roots on disk
733 * The error handling in this function may not be obvious. Any of the
734 * failures will cause the file system to go offline. We still need
735 * to clean up the delayed refs.
737 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
,
738 struct btrfs_root
*root
)
740 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
741 struct list_head
*next
;
742 struct extent_buffer
*eb
;
745 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
749 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
750 ret
= btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
,
752 btrfs_tree_unlock(eb
);
753 free_extent_buffer(eb
);
758 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
762 ret
= btrfs_run_dev_stats(trans
, root
->fs_info
);
765 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
766 next
= fs_info
->dirty_cowonly_roots
.next
;
768 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
770 ret
= update_cowonly_root(trans
, root
);
775 down_write(&fs_info
->extent_commit_sem
);
776 switch_commit_root(fs_info
->extent_root
);
777 up_write(&fs_info
->extent_commit_sem
);
783 * dead roots are old snapshots that need to be deleted. This allocates
784 * a dirty root struct and adds it into the list of dead roots that need to
787 int btrfs_add_dead_root(struct btrfs_root
*root
)
789 spin_lock(&root
->fs_info
->trans_lock
);
790 list_add(&root
->root_list
, &root
->fs_info
->dead_roots
);
791 spin_unlock(&root
->fs_info
->trans_lock
);
796 * update all the cowonly tree roots on disk
798 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
,
799 struct btrfs_root
*root
)
801 struct btrfs_root
*gang
[8];
802 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
807 spin_lock(&fs_info
->fs_roots_radix_lock
);
809 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
812 BTRFS_ROOT_TRANS_TAG
);
815 for (i
= 0; i
< ret
; i
++) {
817 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
818 (unsigned long)root
->root_key
.objectid
,
819 BTRFS_ROOT_TRANS_TAG
);
820 spin_unlock(&fs_info
->fs_roots_radix_lock
);
822 btrfs_free_log(trans
, root
);
823 btrfs_update_reloc_root(trans
, root
);
824 btrfs_orphan_commit_root(trans
, root
);
826 btrfs_save_ino_cache(root
, trans
);
828 /* see comments in should_cow_block() */
832 if (root
->commit_root
!= root
->node
) {
833 mutex_lock(&root
->fs_commit_mutex
);
834 switch_commit_root(root
);
835 btrfs_unpin_free_ino(root
);
836 mutex_unlock(&root
->fs_commit_mutex
);
838 btrfs_set_root_node(&root
->root_item
,
842 err
= btrfs_update_root(trans
, fs_info
->tree_root
,
845 spin_lock(&fs_info
->fs_roots_radix_lock
);
850 spin_unlock(&fs_info
->fs_roots_radix_lock
);
855 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
856 * otherwise every leaf in the btree is read and defragged.
858 int btrfs_defrag_root(struct btrfs_root
*root
, int cacheonly
)
860 struct btrfs_fs_info
*info
= root
->fs_info
;
861 struct btrfs_trans_handle
*trans
;
865 if (xchg(&root
->defrag_running
, 1))
869 trans
= btrfs_start_transaction(root
, 0);
871 return PTR_ERR(trans
);
873 ret
= btrfs_defrag_leaves(trans
, root
, cacheonly
);
875 nr
= trans
->blocks_used
;
876 btrfs_end_transaction(trans
, root
);
877 btrfs_btree_balance_dirty(info
->tree_root
, nr
);
880 if (btrfs_fs_closing(root
->fs_info
) || ret
!= -EAGAIN
)
883 root
->defrag_running
= 0;
888 * new snapshots need to be created at a very specific time in the
889 * transaction commit. This does the actual creation
891 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
892 struct btrfs_fs_info
*fs_info
,
893 struct btrfs_pending_snapshot
*pending
)
895 struct btrfs_key key
;
896 struct btrfs_root_item
*new_root_item
;
897 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
898 struct btrfs_root
*root
= pending
->root
;
899 struct btrfs_root
*parent_root
;
900 struct btrfs_block_rsv
*rsv
;
901 struct inode
*parent_inode
;
902 struct dentry
*parent
;
903 struct dentry
*dentry
;
904 struct extent_buffer
*tmp
;
905 struct extent_buffer
*old
;
912 rsv
= trans
->block_rsv
;
914 new_root_item
= kmalloc(sizeof(*new_root_item
), GFP_NOFS
);
915 if (!new_root_item
) {
916 ret
= pending
->error
= -ENOMEM
;
920 ret
= btrfs_find_free_objectid(tree_root
, &objectid
);
922 pending
->error
= ret
;
926 btrfs_reloc_pre_snapshot(trans
, pending
, &to_reserve
);
928 if (to_reserve
> 0) {
929 ret
= btrfs_block_rsv_add_noflush(root
, &pending
->block_rsv
,
932 pending
->error
= ret
;
937 key
.objectid
= objectid
;
938 key
.offset
= (u64
)-1;
939 key
.type
= BTRFS_ROOT_ITEM_KEY
;
941 trans
->block_rsv
= &pending
->block_rsv
;
943 dentry
= pending
->dentry
;
944 parent
= dget_parent(dentry
);
945 parent_inode
= parent
->d_inode
;
946 parent_root
= BTRFS_I(parent_inode
)->root
;
947 record_root_in_trans(trans
, parent_root
);
950 * insert the directory item
952 ret
= btrfs_set_inode_index(parent_inode
, &index
);
953 BUG_ON(ret
); /* -ENOMEM */
954 ret
= btrfs_insert_dir_item(trans
, parent_root
,
955 dentry
->d_name
.name
, dentry
->d_name
.len
,
957 BTRFS_FT_DIR
, index
);
958 if (ret
== -EEXIST
) {
959 pending
->error
= -EEXIST
;
963 goto abort_trans_dput
;
966 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
967 dentry
->d_name
.len
* 2);
968 ret
= btrfs_update_inode(trans
, parent_root
, parent_inode
);
970 goto abort_trans_dput
;
973 * pull in the delayed directory update
974 * and the delayed inode item
975 * otherwise we corrupt the FS during
978 ret
= btrfs_run_delayed_items(trans
, root
);
979 if (ret
) { /* Transaction aborted */
984 record_root_in_trans(trans
, root
);
985 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
986 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
987 btrfs_check_and_init_root_item(new_root_item
);
989 root_flags
= btrfs_root_flags(new_root_item
);
990 if (pending
->readonly
)
991 root_flags
|= BTRFS_ROOT_SUBVOL_RDONLY
;
993 root_flags
&= ~BTRFS_ROOT_SUBVOL_RDONLY
;
994 btrfs_set_root_flags(new_root_item
, root_flags
);
996 old
= btrfs_lock_root_node(root
);
997 ret
= btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
);
999 btrfs_tree_unlock(old
);
1000 free_extent_buffer(old
);
1001 goto abort_trans_dput
;
1004 btrfs_set_lock_blocking(old
);
1006 ret
= btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
1007 /* clean up in any case */
1008 btrfs_tree_unlock(old
);
1009 free_extent_buffer(old
);
1011 goto abort_trans_dput
;
1013 /* see comments in should_cow_block() */
1014 root
->force_cow
= 1;
1017 btrfs_set_root_node(new_root_item
, tmp
);
1018 /* record when the snapshot was created in key.offset */
1019 key
.offset
= trans
->transid
;
1020 ret
= btrfs_insert_root(trans
, tree_root
, &key
, new_root_item
);
1021 btrfs_tree_unlock(tmp
);
1022 free_extent_buffer(tmp
);
1024 goto abort_trans_dput
;
1027 * insert root back/forward references
1029 ret
= btrfs_add_root_ref(trans
, tree_root
, objectid
,
1030 parent_root
->root_key
.objectid
,
1031 btrfs_ino(parent_inode
), index
,
1032 dentry
->d_name
.name
, dentry
->d_name
.len
);
1037 key
.offset
= (u64
)-1;
1038 pending
->snap
= btrfs_read_fs_root_no_name(root
->fs_info
, &key
);
1039 if (IS_ERR(pending
->snap
)) {
1040 ret
= PTR_ERR(pending
->snap
);
1044 ret
= btrfs_reloc_post_snapshot(trans
, pending
);
1049 kfree(new_root_item
);
1050 trans
->block_rsv
= rsv
;
1051 btrfs_block_rsv_release(root
, &pending
->block_rsv
, (u64
)-1);
1057 btrfs_abort_transaction(trans
, root
, ret
);
1062 * create all the snapshots we've scheduled for creation
1064 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
,
1065 struct btrfs_fs_info
*fs_info
)
1067 struct btrfs_pending_snapshot
*pending
;
1068 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
1070 list_for_each_entry(pending
, head
, list
)
1071 create_pending_snapshot(trans
, fs_info
, pending
);
1075 static void update_super_roots(struct btrfs_root
*root
)
1077 struct btrfs_root_item
*root_item
;
1078 struct btrfs_super_block
*super
;
1080 super
= root
->fs_info
->super_copy
;
1082 root_item
= &root
->fs_info
->chunk_root
->root_item
;
1083 super
->chunk_root
= root_item
->bytenr
;
1084 super
->chunk_root_generation
= root_item
->generation
;
1085 super
->chunk_root_level
= root_item
->level
;
1087 root_item
= &root
->fs_info
->tree_root
->root_item
;
1088 super
->root
= root_item
->bytenr
;
1089 super
->generation
= root_item
->generation
;
1090 super
->root_level
= root_item
->level
;
1091 if (btrfs_test_opt(root
, SPACE_CACHE
))
1092 super
->cache_generation
= root_item
->generation
;
1095 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
1098 spin_lock(&info
->trans_lock
);
1099 if (info
->running_transaction
)
1100 ret
= info
->running_transaction
->in_commit
;
1101 spin_unlock(&info
->trans_lock
);
1105 int btrfs_transaction_blocked(struct btrfs_fs_info
*info
)
1108 spin_lock(&info
->trans_lock
);
1109 if (info
->running_transaction
)
1110 ret
= info
->running_transaction
->blocked
;
1111 spin_unlock(&info
->trans_lock
);
1116 * wait for the current transaction commit to start and block subsequent
1119 static void wait_current_trans_commit_start(struct btrfs_root
*root
,
1120 struct btrfs_transaction
*trans
)
1122 wait_event(root
->fs_info
->transaction_blocked_wait
, trans
->in_commit
);
1126 * wait for the current transaction to start and then become unblocked.
1129 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root
*root
,
1130 struct btrfs_transaction
*trans
)
1132 wait_event(root
->fs_info
->transaction_wait
,
1133 trans
->commit_done
|| (trans
->in_commit
&& !trans
->blocked
));
1137 * commit transactions asynchronously. once btrfs_commit_transaction_async
1138 * returns, any subsequent transaction will not be allowed to join.
1140 struct btrfs_async_commit
{
1141 struct btrfs_trans_handle
*newtrans
;
1142 struct btrfs_root
*root
;
1143 struct delayed_work work
;
1146 static void do_async_commit(struct work_struct
*work
)
1148 struct btrfs_async_commit
*ac
=
1149 container_of(work
, struct btrfs_async_commit
, work
.work
);
1151 btrfs_commit_transaction(ac
->newtrans
, ac
->root
);
1155 int btrfs_commit_transaction_async(struct btrfs_trans_handle
*trans
,
1156 struct btrfs_root
*root
,
1157 int wait_for_unblock
)
1159 struct btrfs_async_commit
*ac
;
1160 struct btrfs_transaction
*cur_trans
;
1162 ac
= kmalloc(sizeof(*ac
), GFP_NOFS
);
1166 INIT_DELAYED_WORK(&ac
->work
, do_async_commit
);
1168 ac
->newtrans
= btrfs_join_transaction(root
);
1169 if (IS_ERR(ac
->newtrans
)) {
1170 int err
= PTR_ERR(ac
->newtrans
);
1175 /* take transaction reference */
1176 cur_trans
= trans
->transaction
;
1177 atomic_inc(&cur_trans
->use_count
);
1179 btrfs_end_transaction(trans
, root
);
1180 schedule_delayed_work(&ac
->work
, 0);
1182 /* wait for transaction to start and unblock */
1183 if (wait_for_unblock
)
1184 wait_current_trans_commit_start_and_unblock(root
, cur_trans
);
1186 wait_current_trans_commit_start(root
, cur_trans
);
1188 if (current
->journal_info
== trans
)
1189 current
->journal_info
= NULL
;
1191 put_transaction(cur_trans
);
1196 static void cleanup_transaction(struct btrfs_trans_handle
*trans
,
1197 struct btrfs_root
*root
)
1199 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1201 WARN_ON(trans
->use_count
> 1);
1203 spin_lock(&root
->fs_info
->trans_lock
);
1204 list_del_init(&cur_trans
->list
);
1205 spin_unlock(&root
->fs_info
->trans_lock
);
1207 btrfs_cleanup_one_transaction(trans
->transaction
, root
);
1209 put_transaction(cur_trans
);
1210 put_transaction(cur_trans
);
1212 trace_btrfs_transaction_commit(root
);
1214 btrfs_scrub_continue(root
);
1216 if (current
->journal_info
== trans
)
1217 current
->journal_info
= NULL
;
1219 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1223 * btrfs_transaction state sequence:
1224 * in_commit = 0, blocked = 0 (initial)
1225 * in_commit = 1, blocked = 1
1229 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
,
1230 struct btrfs_root
*root
)
1232 unsigned long joined
= 0;
1233 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1234 struct btrfs_transaction
*prev_trans
= NULL
;
1237 int should_grow
= 0;
1238 unsigned long now
= get_seconds();
1239 int flush_on_commit
= btrfs_test_opt(root
, FLUSHONCOMMIT
);
1241 btrfs_run_ordered_operations(root
, 0);
1243 btrfs_trans_release_metadata(trans
, root
);
1244 trans
->block_rsv
= NULL
;
1246 if (cur_trans
->aborted
)
1247 goto cleanup_transaction
;
1249 /* make a pass through all the delayed refs we have so far
1250 * any runnings procs may add more while we are here
1252 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1254 goto cleanup_transaction
;
1256 cur_trans
= trans
->transaction
;
1259 * set the flushing flag so procs in this transaction have to
1260 * start sending their work down.
1262 cur_trans
->delayed_refs
.flushing
= 1;
1264 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1266 goto cleanup_transaction
;
1268 spin_lock(&cur_trans
->commit_lock
);
1269 if (cur_trans
->in_commit
) {
1270 spin_unlock(&cur_trans
->commit_lock
);
1271 atomic_inc(&cur_trans
->use_count
);
1272 ret
= btrfs_end_transaction(trans
, root
);
1274 wait_for_commit(root
, cur_trans
);
1276 put_transaction(cur_trans
);
1281 trans
->transaction
->in_commit
= 1;
1282 trans
->transaction
->blocked
= 1;
1283 spin_unlock(&cur_trans
->commit_lock
);
1284 wake_up(&root
->fs_info
->transaction_blocked_wait
);
1286 spin_lock(&root
->fs_info
->trans_lock
);
1287 if (cur_trans
->list
.prev
!= &root
->fs_info
->trans_list
) {
1288 prev_trans
= list_entry(cur_trans
->list
.prev
,
1289 struct btrfs_transaction
, list
);
1290 if (!prev_trans
->commit_done
) {
1291 atomic_inc(&prev_trans
->use_count
);
1292 spin_unlock(&root
->fs_info
->trans_lock
);
1294 wait_for_commit(root
, prev_trans
);
1296 put_transaction(prev_trans
);
1298 spin_unlock(&root
->fs_info
->trans_lock
);
1301 spin_unlock(&root
->fs_info
->trans_lock
);
1304 if (now
< cur_trans
->start_time
|| now
- cur_trans
->start_time
< 1)
1308 int snap_pending
= 0;
1310 joined
= cur_trans
->num_joined
;
1311 if (!list_empty(&trans
->transaction
->pending_snapshots
))
1314 WARN_ON(cur_trans
!= trans
->transaction
);
1316 if (flush_on_commit
|| snap_pending
) {
1317 btrfs_start_delalloc_inodes(root
, 1);
1318 btrfs_wait_ordered_extents(root
, 0, 1);
1321 ret
= btrfs_run_delayed_items(trans
, root
);
1323 goto cleanup_transaction
;
1326 * rename don't use btrfs_join_transaction, so, once we
1327 * set the transaction to blocked above, we aren't going
1328 * to get any new ordered operations. We can safely run
1329 * it here and no for sure that nothing new will be added
1332 btrfs_run_ordered_operations(root
, 1);
1334 prepare_to_wait(&cur_trans
->writer_wait
, &wait
,
1335 TASK_UNINTERRUPTIBLE
);
1337 if (atomic_read(&cur_trans
->num_writers
) > 1)
1338 schedule_timeout(MAX_SCHEDULE_TIMEOUT
);
1339 else if (should_grow
)
1340 schedule_timeout(1);
1342 finish_wait(&cur_trans
->writer_wait
, &wait
);
1343 } while (atomic_read(&cur_trans
->num_writers
) > 1 ||
1344 (should_grow
&& cur_trans
->num_joined
!= joined
));
1347 * Ok now we need to make sure to block out any other joins while we
1348 * commit the transaction. We could have started a join before setting
1349 * no_join so make sure to wait for num_writers to == 1 again.
1351 spin_lock(&root
->fs_info
->trans_lock
);
1352 root
->fs_info
->trans_no_join
= 1;
1353 spin_unlock(&root
->fs_info
->trans_lock
);
1354 wait_event(cur_trans
->writer_wait
,
1355 atomic_read(&cur_trans
->num_writers
) == 1);
1358 * the reloc mutex makes sure that we stop
1359 * the balancing code from coming in and moving
1360 * extents around in the middle of the commit
1362 mutex_lock(&root
->fs_info
->reloc_mutex
);
1364 ret
= btrfs_run_delayed_items(trans
, root
);
1366 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1367 goto cleanup_transaction
;
1370 ret
= create_pending_snapshots(trans
, root
->fs_info
);
1372 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1373 goto cleanup_transaction
;
1376 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1378 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1379 goto cleanup_transaction
;
1383 * make sure none of the code above managed to slip in a
1386 btrfs_assert_delayed_root_empty(root
);
1388 WARN_ON(cur_trans
!= trans
->transaction
);
1390 btrfs_scrub_pause(root
);
1391 /* btrfs_commit_tree_roots is responsible for getting the
1392 * various roots consistent with each other. Every pointer
1393 * in the tree of tree roots has to point to the most up to date
1394 * root for every subvolume and other tree. So, we have to keep
1395 * the tree logging code from jumping in and changing any
1398 * At this point in the commit, there can't be any tree-log
1399 * writers, but a little lower down we drop the trans mutex
1400 * and let new people in. By holding the tree_log_mutex
1401 * from now until after the super is written, we avoid races
1402 * with the tree-log code.
1404 mutex_lock(&root
->fs_info
->tree_log_mutex
);
1406 ret
= commit_fs_roots(trans
, root
);
1408 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1409 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1410 goto cleanup_transaction
;
1413 /* commit_fs_roots gets rid of all the tree log roots, it is now
1414 * safe to free the root of tree log roots
1416 btrfs_free_log_root_tree(trans
, root
->fs_info
);
1418 ret
= commit_cowonly_roots(trans
, root
);
1420 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1421 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1422 goto cleanup_transaction
;
1425 btrfs_prepare_extent_commit(trans
, root
);
1427 cur_trans
= root
->fs_info
->running_transaction
;
1429 btrfs_set_root_node(&root
->fs_info
->tree_root
->root_item
,
1430 root
->fs_info
->tree_root
->node
);
1431 switch_commit_root(root
->fs_info
->tree_root
);
1433 btrfs_set_root_node(&root
->fs_info
->chunk_root
->root_item
,
1434 root
->fs_info
->chunk_root
->node
);
1435 switch_commit_root(root
->fs_info
->chunk_root
);
1437 update_super_roots(root
);
1439 if (!root
->fs_info
->log_root_recovering
) {
1440 btrfs_set_super_log_root(root
->fs_info
->super_copy
, 0);
1441 btrfs_set_super_log_root_level(root
->fs_info
->super_copy
, 0);
1444 memcpy(root
->fs_info
->super_for_commit
, root
->fs_info
->super_copy
,
1445 sizeof(*root
->fs_info
->super_copy
));
1447 trans
->transaction
->blocked
= 0;
1448 spin_lock(&root
->fs_info
->trans_lock
);
1449 root
->fs_info
->running_transaction
= NULL
;
1450 root
->fs_info
->trans_no_join
= 0;
1451 spin_unlock(&root
->fs_info
->trans_lock
);
1452 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1454 wake_up(&root
->fs_info
->transaction_wait
);
1456 ret
= btrfs_write_and_wait_transaction(trans
, root
);
1458 btrfs_error(root
->fs_info
, ret
,
1459 "Error while writing out transaction.");
1460 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1461 goto cleanup_transaction
;
1464 ret
= write_ctree_super(trans
, root
, 0);
1466 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1467 goto cleanup_transaction
;
1471 * the super is written, we can safely allow the tree-loggers
1472 * to go about their business
1474 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1476 btrfs_finish_extent_commit(trans
, root
);
1478 cur_trans
->commit_done
= 1;
1480 root
->fs_info
->last_trans_committed
= cur_trans
->transid
;
1482 wake_up(&cur_trans
->commit_wait
);
1484 spin_lock(&root
->fs_info
->trans_lock
);
1485 list_del_init(&cur_trans
->list
);
1486 spin_unlock(&root
->fs_info
->trans_lock
);
1488 put_transaction(cur_trans
);
1489 put_transaction(cur_trans
);
1491 trace_btrfs_transaction_commit(root
);
1493 btrfs_scrub_continue(root
);
1495 if (current
->journal_info
== trans
)
1496 current
->journal_info
= NULL
;
1498 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1500 if (current
!= root
->fs_info
->transaction_kthread
)
1501 btrfs_run_delayed_iputs(root
);
1505 cleanup_transaction
:
1506 btrfs_printk(root
->fs_info
, "Skipping commit of aborted transaction.\n");
1508 if (current
->journal_info
== trans
)
1509 current
->journal_info
= NULL
;
1510 cleanup_transaction(trans
, root
);
1516 * interface function to delete all the snapshots we have scheduled for deletion
1518 int btrfs_clean_old_snapshots(struct btrfs_root
*root
)
1521 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1523 spin_lock(&fs_info
->trans_lock
);
1524 list_splice_init(&fs_info
->dead_roots
, &list
);
1525 spin_unlock(&fs_info
->trans_lock
);
1527 while (!list_empty(&list
)) {
1530 root
= list_entry(list
.next
, struct btrfs_root
, root_list
);
1531 list_del(&root
->root_list
);
1533 btrfs_kill_all_delayed_nodes(root
);
1535 if (btrfs_header_backref_rev(root
->node
) <
1536 BTRFS_MIXED_BACKREF_REV
)
1537 ret
= btrfs_drop_snapshot(root
, NULL
, 0, 0);
1539 ret
=btrfs_drop_snapshot(root
, NULL
, 1, 0);