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"
32 #define BTRFS_ROOT_TRANS_TAG 0
34 static noinline
void put_transaction(struct btrfs_transaction
*transaction
)
36 WARN_ON(atomic_read(&transaction
->use_count
) == 0);
37 if (atomic_dec_and_test(&transaction
->use_count
)) {
38 BUG_ON(!list_empty(&transaction
->list
));
39 WARN_ON(transaction
->delayed_refs
.root
.rb_node
);
40 WARN_ON(!list_empty(&transaction
->delayed_refs
.seq_head
));
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
;
59 spin_lock(&root
->fs_info
->trans_lock
);
61 if (root
->fs_info
->trans_no_join
) {
63 spin_unlock(&root
->fs_info
->trans_lock
);
68 cur_trans
= root
->fs_info
->running_transaction
;
70 atomic_inc(&cur_trans
->use_count
);
71 atomic_inc(&cur_trans
->num_writers
);
72 cur_trans
->num_joined
++;
73 spin_unlock(&root
->fs_info
->trans_lock
);
76 spin_unlock(&root
->fs_info
->trans_lock
);
78 cur_trans
= kmem_cache_alloc(btrfs_transaction_cachep
, GFP_NOFS
);
82 spin_lock(&root
->fs_info
->trans_lock
);
83 if (root
->fs_info
->running_transaction
) {
85 * someone started a transaction after we unlocked. Make sure
86 * to redo the trans_no_join checks above
88 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
89 cur_trans
= root
->fs_info
->running_transaction
;
93 atomic_set(&cur_trans
->num_writers
, 1);
94 cur_trans
->num_joined
= 0;
95 init_waitqueue_head(&cur_trans
->writer_wait
);
96 init_waitqueue_head(&cur_trans
->commit_wait
);
97 cur_trans
->in_commit
= 0;
98 cur_trans
->blocked
= 0;
100 * One for this trans handle, one so it will live on until we
101 * commit the transaction.
103 atomic_set(&cur_trans
->use_count
, 2);
104 cur_trans
->commit_done
= 0;
105 cur_trans
->start_time
= get_seconds();
107 cur_trans
->delayed_refs
.root
= RB_ROOT
;
108 cur_trans
->delayed_refs
.num_entries
= 0;
109 cur_trans
->delayed_refs
.num_heads_ready
= 0;
110 cur_trans
->delayed_refs
.num_heads
= 0;
111 cur_trans
->delayed_refs
.flushing
= 0;
112 cur_trans
->delayed_refs
.run_delayed_start
= 0;
113 cur_trans
->delayed_refs
.seq
= 1;
114 init_waitqueue_head(&cur_trans
->delayed_refs
.seq_wait
);
115 spin_lock_init(&cur_trans
->commit_lock
);
116 spin_lock_init(&cur_trans
->delayed_refs
.lock
);
117 INIT_LIST_HEAD(&cur_trans
->delayed_refs
.seq_head
);
119 INIT_LIST_HEAD(&cur_trans
->pending_snapshots
);
120 list_add_tail(&cur_trans
->list
, &root
->fs_info
->trans_list
);
121 extent_io_tree_init(&cur_trans
->dirty_pages
,
122 root
->fs_info
->btree_inode
->i_mapping
);
123 root
->fs_info
->generation
++;
124 cur_trans
->transid
= root
->fs_info
->generation
;
125 root
->fs_info
->running_transaction
= cur_trans
;
126 spin_unlock(&root
->fs_info
->trans_lock
);
132 * this does all the record keeping required to make sure that a reference
133 * counted root is properly recorded in a given transaction. This is required
134 * to make sure the old root from before we joined the transaction is deleted
135 * when the transaction commits
137 static int record_root_in_trans(struct btrfs_trans_handle
*trans
,
138 struct btrfs_root
*root
)
140 if (root
->ref_cows
&& root
->last_trans
< trans
->transid
) {
141 WARN_ON(root
== root
->fs_info
->extent_root
);
142 WARN_ON(root
->commit_root
!= root
->node
);
145 * see below for in_trans_setup usage rules
146 * we have the reloc mutex held now, so there
147 * is only one writer in this function
149 root
->in_trans_setup
= 1;
151 /* make sure readers find in_trans_setup before
152 * they find our root->last_trans update
156 spin_lock(&root
->fs_info
->fs_roots_radix_lock
);
157 if (root
->last_trans
== trans
->transid
) {
158 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
161 radix_tree_tag_set(&root
->fs_info
->fs_roots_radix
,
162 (unsigned long)root
->root_key
.objectid
,
163 BTRFS_ROOT_TRANS_TAG
);
164 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
165 root
->last_trans
= trans
->transid
;
167 /* this is pretty tricky. We don't want to
168 * take the relocation lock in btrfs_record_root_in_trans
169 * unless we're really doing the first setup for this root in
172 * Normally we'd use root->last_trans as a flag to decide
173 * if we want to take the expensive mutex.
175 * But, we have to set root->last_trans before we
176 * init the relocation root, otherwise, we trip over warnings
177 * in ctree.c. The solution used here is to flag ourselves
178 * with root->in_trans_setup. When this is 1, we're still
179 * fixing up the reloc trees and everyone must wait.
181 * When this is zero, they can trust root->last_trans and fly
182 * through btrfs_record_root_in_trans without having to take the
183 * lock. smp_wmb() makes sure that all the writes above are
184 * done before we pop in the zero below
186 btrfs_init_reloc_root(trans
, root
);
188 root
->in_trans_setup
= 0;
194 int btrfs_record_root_in_trans(struct btrfs_trans_handle
*trans
,
195 struct btrfs_root
*root
)
201 * see record_root_in_trans for comments about in_trans_setup usage
205 if (root
->last_trans
== trans
->transid
&&
206 !root
->in_trans_setup
)
209 mutex_lock(&root
->fs_info
->reloc_mutex
);
210 record_root_in_trans(trans
, root
);
211 mutex_unlock(&root
->fs_info
->reloc_mutex
);
216 /* wait for commit against the current transaction to become unblocked
217 * when this is done, it is safe to start a new transaction, but the current
218 * transaction might not be fully on disk.
220 static void wait_current_trans(struct btrfs_root
*root
)
222 struct btrfs_transaction
*cur_trans
;
224 spin_lock(&root
->fs_info
->trans_lock
);
225 cur_trans
= root
->fs_info
->running_transaction
;
226 if (cur_trans
&& cur_trans
->blocked
) {
227 atomic_inc(&cur_trans
->use_count
);
228 spin_unlock(&root
->fs_info
->trans_lock
);
230 wait_event(root
->fs_info
->transaction_wait
,
231 !cur_trans
->blocked
);
232 put_transaction(cur_trans
);
234 spin_unlock(&root
->fs_info
->trans_lock
);
238 enum btrfs_trans_type
{
245 static int may_wait_transaction(struct btrfs_root
*root
, int type
)
247 if (root
->fs_info
->log_root_recovering
)
250 if (type
== TRANS_USERSPACE
)
253 if (type
== TRANS_START
&&
254 !atomic_read(&root
->fs_info
->open_ioctl_trans
))
260 static struct btrfs_trans_handle
*start_transaction(struct btrfs_root
*root
,
261 u64 num_items
, int type
)
263 struct btrfs_trans_handle
*h
;
264 struct btrfs_transaction
*cur_trans
;
268 if (root
->fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
)
269 return ERR_PTR(-EROFS
);
271 if (current
->journal_info
) {
272 WARN_ON(type
!= TRANS_JOIN
&& type
!= TRANS_JOIN_NOLOCK
);
273 h
= current
->journal_info
;
275 h
->orig_rsv
= h
->block_rsv
;
281 * Do the reservation before we join the transaction so we can do all
282 * the appropriate flushing if need be.
284 if (num_items
> 0 && root
!= root
->fs_info
->chunk_root
) {
285 num_bytes
= btrfs_calc_trans_metadata_size(root
, num_items
);
286 ret
= btrfs_block_rsv_add(root
,
287 &root
->fs_info
->trans_block_rsv
,
293 h
= kmem_cache_alloc(btrfs_trans_handle_cachep
, GFP_NOFS
);
295 return ERR_PTR(-ENOMEM
);
297 if (may_wait_transaction(root
, type
))
298 wait_current_trans(root
);
301 ret
= join_transaction(root
, type
== TRANS_JOIN_NOLOCK
);
303 wait_current_trans(root
);
304 } while (ret
== -EBUSY
);
307 kmem_cache_free(btrfs_trans_handle_cachep
, h
);
311 cur_trans
= root
->fs_info
->running_transaction
;
313 h
->transid
= cur_trans
->transid
;
314 h
->transaction
= cur_trans
;
316 h
->bytes_reserved
= 0;
317 h
->delayed_ref_updates
= 0;
323 if (cur_trans
->blocked
&& may_wait_transaction(root
, type
)) {
324 btrfs_commit_transaction(h
, root
);
329 h
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
330 h
->bytes_reserved
= num_bytes
;
334 btrfs_record_root_in_trans(h
, root
);
336 if (!current
->journal_info
&& type
!= TRANS_USERSPACE
)
337 current
->journal_info
= h
;
341 struct btrfs_trans_handle
*btrfs_start_transaction(struct btrfs_root
*root
,
344 return start_transaction(root
, num_items
, TRANS_START
);
346 struct btrfs_trans_handle
*btrfs_join_transaction(struct btrfs_root
*root
)
348 return start_transaction(root
, 0, TRANS_JOIN
);
351 struct btrfs_trans_handle
*btrfs_join_transaction_nolock(struct btrfs_root
*root
)
353 return start_transaction(root
, 0, TRANS_JOIN_NOLOCK
);
356 struct btrfs_trans_handle
*btrfs_start_ioctl_transaction(struct btrfs_root
*root
)
358 return start_transaction(root
, 0, TRANS_USERSPACE
);
361 /* wait for a transaction commit to be fully complete */
362 static noinline
void wait_for_commit(struct btrfs_root
*root
,
363 struct btrfs_transaction
*commit
)
365 wait_event(commit
->commit_wait
, commit
->commit_done
);
368 int btrfs_wait_for_commit(struct btrfs_root
*root
, u64 transid
)
370 struct btrfs_transaction
*cur_trans
= NULL
, *t
;
375 if (transid
<= root
->fs_info
->last_trans_committed
)
378 /* find specified transaction */
379 spin_lock(&root
->fs_info
->trans_lock
);
380 list_for_each_entry(t
, &root
->fs_info
->trans_list
, list
) {
381 if (t
->transid
== transid
) {
383 atomic_inc(&cur_trans
->use_count
);
386 if (t
->transid
> transid
)
389 spin_unlock(&root
->fs_info
->trans_lock
);
392 goto out
; /* bad transid */
394 /* find newest transaction that is committing | committed */
395 spin_lock(&root
->fs_info
->trans_lock
);
396 list_for_each_entry_reverse(t
, &root
->fs_info
->trans_list
,
402 atomic_inc(&cur_trans
->use_count
);
406 spin_unlock(&root
->fs_info
->trans_lock
);
408 goto out
; /* nothing committing|committed */
411 wait_for_commit(root
, cur_trans
);
413 put_transaction(cur_trans
);
419 void btrfs_throttle(struct btrfs_root
*root
)
421 if (!atomic_read(&root
->fs_info
->open_ioctl_trans
))
422 wait_current_trans(root
);
425 static int should_end_transaction(struct btrfs_trans_handle
*trans
,
426 struct btrfs_root
*root
)
430 ret
= btrfs_block_rsv_check(root
, &root
->fs_info
->global_block_rsv
, 5);
434 int btrfs_should_end_transaction(struct btrfs_trans_handle
*trans
,
435 struct btrfs_root
*root
)
437 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
438 struct btrfs_block_rsv
*rsv
= trans
->block_rsv
;
442 if (cur_trans
->blocked
|| cur_trans
->delayed_refs
.flushing
)
446 * We need to do this in case we're deleting csums so the global block
447 * rsv get's used instead of the csum block rsv.
449 trans
->block_rsv
= NULL
;
451 updates
= trans
->delayed_ref_updates
;
452 trans
->delayed_ref_updates
= 0;
454 btrfs_run_delayed_refs(trans
, root
, updates
);
456 trans
->block_rsv
= rsv
;
458 return should_end_transaction(trans
, root
);
461 static int __btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
462 struct btrfs_root
*root
, int throttle
, int lock
)
464 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
465 struct btrfs_fs_info
*info
= root
->fs_info
;
468 if (--trans
->use_count
) {
469 trans
->block_rsv
= trans
->orig_rsv
;
473 btrfs_trans_release_metadata(trans
, root
);
474 trans
->block_rsv
= NULL
;
476 unsigned long cur
= trans
->delayed_ref_updates
;
477 trans
->delayed_ref_updates
= 0;
479 trans
->transaction
->delayed_refs
.num_heads_ready
> 64) {
480 trans
->delayed_ref_updates
= 0;
481 btrfs_run_delayed_refs(trans
, root
, cur
);
488 if (lock
&& !atomic_read(&root
->fs_info
->open_ioctl_trans
) &&
489 should_end_transaction(trans
, root
)) {
490 trans
->transaction
->blocked
= 1;
494 if (lock
&& cur_trans
->blocked
&& !cur_trans
->in_commit
) {
497 * We may race with somebody else here so end up having
498 * to call end_transaction on ourselves again, so inc
502 return btrfs_commit_transaction(trans
, root
);
504 wake_up_process(info
->transaction_kthread
);
508 WARN_ON(cur_trans
!= info
->running_transaction
);
509 WARN_ON(atomic_read(&cur_trans
->num_writers
) < 1);
510 atomic_dec(&cur_trans
->num_writers
);
513 if (waitqueue_active(&cur_trans
->writer_wait
))
514 wake_up(&cur_trans
->writer_wait
);
515 put_transaction(cur_trans
);
517 if (current
->journal_info
== trans
)
518 current
->journal_info
= NULL
;
519 memset(trans
, 0, sizeof(*trans
));
520 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
523 btrfs_run_delayed_iputs(root
);
528 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
529 struct btrfs_root
*root
)
533 ret
= __btrfs_end_transaction(trans
, root
, 0, 1);
539 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
,
540 struct btrfs_root
*root
)
544 ret
= __btrfs_end_transaction(trans
, root
, 1, 1);
550 int btrfs_end_transaction_nolock(struct btrfs_trans_handle
*trans
,
551 struct btrfs_root
*root
)
555 ret
= __btrfs_end_transaction(trans
, root
, 0, 0);
561 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle
*trans
,
562 struct btrfs_root
*root
)
564 return __btrfs_end_transaction(trans
, root
, 1, 1);
568 * when btree blocks are allocated, they have some corresponding bits set for
569 * them in one of two extent_io trees. This is used to make sure all of
570 * those extents are sent to disk but does not wait on them
572 int btrfs_write_marked_extents(struct btrfs_root
*root
,
573 struct extent_io_tree
*dirty_pages
, int mark
)
577 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
581 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
583 convert_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
, mark
,
585 err
= filemap_fdatawrite_range(mapping
, start
, end
);
597 * when btree blocks are allocated, they have some corresponding bits set for
598 * them in one of two extent_io trees. This is used to make sure all of
599 * those extents are on disk for transaction or log commit. We wait
600 * on all the pages and clear them from the dirty pages state tree
602 int btrfs_wait_marked_extents(struct btrfs_root
*root
,
603 struct extent_io_tree
*dirty_pages
, int mark
)
607 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
611 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
613 clear_extent_bits(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
, GFP_NOFS
);
614 err
= filemap_fdatawait_range(mapping
, start
, end
);
626 * when btree blocks are allocated, they have some corresponding bits set for
627 * them in one of two extent_io trees. This is used to make sure all of
628 * those extents are on disk for transaction or log commit
630 int btrfs_write_and_wait_marked_extents(struct btrfs_root
*root
,
631 struct extent_io_tree
*dirty_pages
, int mark
)
636 ret
= btrfs_write_marked_extents(root
, dirty_pages
, mark
);
637 ret2
= btrfs_wait_marked_extents(root
, dirty_pages
, mark
);
646 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
,
647 struct btrfs_root
*root
)
649 if (!trans
|| !trans
->transaction
) {
650 struct inode
*btree_inode
;
651 btree_inode
= root
->fs_info
->btree_inode
;
652 return filemap_write_and_wait(btree_inode
->i_mapping
);
654 return btrfs_write_and_wait_marked_extents(root
,
655 &trans
->transaction
->dirty_pages
,
660 * this is used to update the root pointer in the tree of tree roots.
662 * But, in the case of the extent allocation tree, updating the root
663 * pointer may allocate blocks which may change the root of the extent
666 * So, this loops and repeats and makes sure the cowonly root didn't
667 * change while the root pointer was being updated in the metadata.
669 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
670 struct btrfs_root
*root
)
675 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
677 old_root_used
= btrfs_root_used(&root
->root_item
);
678 btrfs_write_dirty_block_groups(trans
, root
);
681 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
682 if (old_root_bytenr
== root
->node
->start
&&
683 old_root_used
== btrfs_root_used(&root
->root_item
))
686 btrfs_set_root_node(&root
->root_item
, root
->node
);
687 ret
= btrfs_update_root(trans
, tree_root
,
692 old_root_used
= btrfs_root_used(&root
->root_item
);
693 ret
= btrfs_write_dirty_block_groups(trans
, root
);
697 if (root
!= root
->fs_info
->extent_root
)
698 switch_commit_root(root
);
704 * update all the cowonly tree roots on disk
706 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
,
707 struct btrfs_root
*root
)
709 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
710 struct list_head
*next
;
711 struct extent_buffer
*eb
;
714 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
717 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
718 btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
, 0, &eb
);
719 btrfs_tree_unlock(eb
);
720 free_extent_buffer(eb
);
722 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
725 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
726 next
= fs_info
->dirty_cowonly_roots
.next
;
728 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
730 update_cowonly_root(trans
, root
);
733 down_write(&fs_info
->extent_commit_sem
);
734 switch_commit_root(fs_info
->extent_root
);
735 up_write(&fs_info
->extent_commit_sem
);
741 * dead roots are old snapshots that need to be deleted. This allocates
742 * a dirty root struct and adds it into the list of dead roots that need to
745 int btrfs_add_dead_root(struct btrfs_root
*root
)
747 spin_lock(&root
->fs_info
->trans_lock
);
748 list_add(&root
->root_list
, &root
->fs_info
->dead_roots
);
749 spin_unlock(&root
->fs_info
->trans_lock
);
754 * update all the cowonly tree roots on disk
756 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
,
757 struct btrfs_root
*root
)
759 struct btrfs_root
*gang
[8];
760 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
765 spin_lock(&fs_info
->fs_roots_radix_lock
);
767 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
770 BTRFS_ROOT_TRANS_TAG
);
773 for (i
= 0; i
< ret
; i
++) {
775 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
776 (unsigned long)root
->root_key
.objectid
,
777 BTRFS_ROOT_TRANS_TAG
);
778 spin_unlock(&fs_info
->fs_roots_radix_lock
);
780 btrfs_free_log(trans
, root
);
781 btrfs_update_reloc_root(trans
, root
);
782 btrfs_orphan_commit_root(trans
, root
);
784 btrfs_save_ino_cache(root
, trans
);
786 /* see comments in should_cow_block() */
790 if (root
->commit_root
!= root
->node
) {
791 mutex_lock(&root
->fs_commit_mutex
);
792 switch_commit_root(root
);
793 btrfs_unpin_free_ino(root
);
794 mutex_unlock(&root
->fs_commit_mutex
);
796 btrfs_set_root_node(&root
->root_item
,
800 err
= btrfs_update_root(trans
, fs_info
->tree_root
,
803 spin_lock(&fs_info
->fs_roots_radix_lock
);
808 spin_unlock(&fs_info
->fs_roots_radix_lock
);
813 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
814 * otherwise every leaf in the btree is read and defragged.
816 int btrfs_defrag_root(struct btrfs_root
*root
, int cacheonly
)
818 struct btrfs_fs_info
*info
= root
->fs_info
;
819 struct btrfs_trans_handle
*trans
;
823 if (xchg(&root
->defrag_running
, 1))
827 trans
= btrfs_start_transaction(root
, 0);
829 return PTR_ERR(trans
);
831 ret
= btrfs_defrag_leaves(trans
, root
, cacheonly
);
833 nr
= trans
->blocks_used
;
834 btrfs_end_transaction(trans
, root
);
835 btrfs_btree_balance_dirty(info
->tree_root
, nr
);
838 if (btrfs_fs_closing(root
->fs_info
) || ret
!= -EAGAIN
)
841 root
->defrag_running
= 0;
846 * new snapshots need to be created at a very specific time in the
847 * transaction commit. This does the actual creation
849 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
850 struct btrfs_fs_info
*fs_info
,
851 struct btrfs_pending_snapshot
*pending
)
853 struct btrfs_key key
;
854 struct btrfs_root_item
*new_root_item
;
855 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
856 struct btrfs_root
*root
= pending
->root
;
857 struct btrfs_root
*parent_root
;
858 struct btrfs_block_rsv
*rsv
;
859 struct inode
*parent_inode
;
860 struct dentry
*parent
;
861 struct dentry
*dentry
;
862 struct extent_buffer
*tmp
;
863 struct extent_buffer
*old
;
870 rsv
= trans
->block_rsv
;
872 new_root_item
= kmalloc(sizeof(*new_root_item
), GFP_NOFS
);
873 if (!new_root_item
) {
874 pending
->error
= -ENOMEM
;
878 ret
= btrfs_find_free_objectid(tree_root
, &objectid
);
880 pending
->error
= ret
;
884 btrfs_reloc_pre_snapshot(trans
, pending
, &to_reserve
);
886 if (to_reserve
> 0) {
887 ret
= btrfs_block_rsv_add_noflush(root
, &pending
->block_rsv
,
890 pending
->error
= ret
;
895 key
.objectid
= objectid
;
896 key
.offset
= (u64
)-1;
897 key
.type
= BTRFS_ROOT_ITEM_KEY
;
899 trans
->block_rsv
= &pending
->block_rsv
;
901 dentry
= pending
->dentry
;
902 parent
= dget_parent(dentry
);
903 parent_inode
= parent
->d_inode
;
904 parent_root
= BTRFS_I(parent_inode
)->root
;
905 record_root_in_trans(trans
, parent_root
);
908 * insert the directory item
910 ret
= btrfs_set_inode_index(parent_inode
, &index
);
912 ret
= btrfs_insert_dir_item(trans
, parent_root
,
913 dentry
->d_name
.name
, dentry
->d_name
.len
,
915 BTRFS_FT_DIR
, index
);
918 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
919 dentry
->d_name
.len
* 2);
920 ret
= btrfs_update_inode(trans
, parent_root
, parent_inode
);
924 * pull in the delayed directory update
925 * and the delayed inode item
926 * otherwise we corrupt the FS during
929 ret
= btrfs_run_delayed_items(trans
, root
);
932 record_root_in_trans(trans
, root
);
933 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
934 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
935 btrfs_check_and_init_root_item(new_root_item
);
937 root_flags
= btrfs_root_flags(new_root_item
);
938 if (pending
->readonly
)
939 root_flags
|= BTRFS_ROOT_SUBVOL_RDONLY
;
941 root_flags
&= ~BTRFS_ROOT_SUBVOL_RDONLY
;
942 btrfs_set_root_flags(new_root_item
, root_flags
);
944 old
= btrfs_lock_root_node(root
);
945 btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
);
946 btrfs_set_lock_blocking(old
);
948 btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
949 btrfs_tree_unlock(old
);
950 free_extent_buffer(old
);
952 /* see comments in should_cow_block() */
956 btrfs_set_root_node(new_root_item
, tmp
);
957 /* record when the snapshot was created in key.offset */
958 key
.offset
= trans
->transid
;
959 ret
= btrfs_insert_root(trans
, tree_root
, &key
, new_root_item
);
960 btrfs_tree_unlock(tmp
);
961 free_extent_buffer(tmp
);
965 * insert root back/forward references
967 ret
= btrfs_add_root_ref(trans
, tree_root
, objectid
,
968 parent_root
->root_key
.objectid
,
969 btrfs_ino(parent_inode
), index
,
970 dentry
->d_name
.name
, dentry
->d_name
.len
);
974 key
.offset
= (u64
)-1;
975 pending
->snap
= btrfs_read_fs_root_no_name(root
->fs_info
, &key
);
976 BUG_ON(IS_ERR(pending
->snap
));
978 btrfs_reloc_post_snapshot(trans
, pending
);
980 kfree(new_root_item
);
981 trans
->block_rsv
= rsv
;
982 btrfs_block_rsv_release(root
, &pending
->block_rsv
, (u64
)-1);
987 * create all the snapshots we've scheduled for creation
989 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
,
990 struct btrfs_fs_info
*fs_info
)
992 struct btrfs_pending_snapshot
*pending
;
993 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
996 list_for_each_entry(pending
, head
, list
) {
997 ret
= create_pending_snapshot(trans
, fs_info
, pending
);
1003 static void update_super_roots(struct btrfs_root
*root
)
1005 struct btrfs_root_item
*root_item
;
1006 struct btrfs_super_block
*super
;
1008 super
= root
->fs_info
->super_copy
;
1010 root_item
= &root
->fs_info
->chunk_root
->root_item
;
1011 super
->chunk_root
= root_item
->bytenr
;
1012 super
->chunk_root_generation
= root_item
->generation
;
1013 super
->chunk_root_level
= root_item
->level
;
1015 root_item
= &root
->fs_info
->tree_root
->root_item
;
1016 super
->root
= root_item
->bytenr
;
1017 super
->generation
= root_item
->generation
;
1018 super
->root_level
= root_item
->level
;
1019 if (btrfs_test_opt(root
, SPACE_CACHE
))
1020 super
->cache_generation
= root_item
->generation
;
1023 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
1026 spin_lock(&info
->trans_lock
);
1027 if (info
->running_transaction
)
1028 ret
= info
->running_transaction
->in_commit
;
1029 spin_unlock(&info
->trans_lock
);
1033 int btrfs_transaction_blocked(struct btrfs_fs_info
*info
)
1036 spin_lock(&info
->trans_lock
);
1037 if (info
->running_transaction
)
1038 ret
= info
->running_transaction
->blocked
;
1039 spin_unlock(&info
->trans_lock
);
1044 * wait for the current transaction commit to start and block subsequent
1047 static void wait_current_trans_commit_start(struct btrfs_root
*root
,
1048 struct btrfs_transaction
*trans
)
1050 wait_event(root
->fs_info
->transaction_blocked_wait
, trans
->in_commit
);
1054 * wait for the current transaction to start and then become unblocked.
1057 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root
*root
,
1058 struct btrfs_transaction
*trans
)
1060 wait_event(root
->fs_info
->transaction_wait
,
1061 trans
->commit_done
|| (trans
->in_commit
&& !trans
->blocked
));
1065 * commit transactions asynchronously. once btrfs_commit_transaction_async
1066 * returns, any subsequent transaction will not be allowed to join.
1068 struct btrfs_async_commit
{
1069 struct btrfs_trans_handle
*newtrans
;
1070 struct btrfs_root
*root
;
1071 struct delayed_work work
;
1074 static void do_async_commit(struct work_struct
*work
)
1076 struct btrfs_async_commit
*ac
=
1077 container_of(work
, struct btrfs_async_commit
, work
.work
);
1079 btrfs_commit_transaction(ac
->newtrans
, ac
->root
);
1083 int btrfs_commit_transaction_async(struct btrfs_trans_handle
*trans
,
1084 struct btrfs_root
*root
,
1085 int wait_for_unblock
)
1087 struct btrfs_async_commit
*ac
;
1088 struct btrfs_transaction
*cur_trans
;
1090 ac
= kmalloc(sizeof(*ac
), GFP_NOFS
);
1094 INIT_DELAYED_WORK(&ac
->work
, do_async_commit
);
1096 ac
->newtrans
= btrfs_join_transaction(root
);
1097 if (IS_ERR(ac
->newtrans
)) {
1098 int err
= PTR_ERR(ac
->newtrans
);
1103 /* take transaction reference */
1104 cur_trans
= trans
->transaction
;
1105 atomic_inc(&cur_trans
->use_count
);
1107 btrfs_end_transaction(trans
, root
);
1108 schedule_delayed_work(&ac
->work
, 0);
1110 /* wait for transaction to start and unblock */
1111 if (wait_for_unblock
)
1112 wait_current_trans_commit_start_and_unblock(root
, cur_trans
);
1114 wait_current_trans_commit_start(root
, cur_trans
);
1116 if (current
->journal_info
== trans
)
1117 current
->journal_info
= NULL
;
1119 put_transaction(cur_trans
);
1124 * btrfs_transaction state sequence:
1125 * in_commit = 0, blocked = 0 (initial)
1126 * in_commit = 1, blocked = 1
1130 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
,
1131 struct btrfs_root
*root
)
1133 unsigned long joined
= 0;
1134 struct btrfs_transaction
*cur_trans
;
1135 struct btrfs_transaction
*prev_trans
= NULL
;
1138 int should_grow
= 0;
1139 unsigned long now
= get_seconds();
1140 int flush_on_commit
= btrfs_test_opt(root
, FLUSHONCOMMIT
);
1142 btrfs_run_ordered_operations(root
, 0);
1144 btrfs_trans_release_metadata(trans
, root
);
1145 trans
->block_rsv
= NULL
;
1147 /* make a pass through all the delayed refs we have so far
1148 * any runnings procs may add more while we are here
1150 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1153 cur_trans
= trans
->transaction
;
1155 * set the flushing flag so procs in this transaction have to
1156 * start sending their work down.
1158 cur_trans
->delayed_refs
.flushing
= 1;
1160 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1163 spin_lock(&cur_trans
->commit_lock
);
1164 if (cur_trans
->in_commit
) {
1165 spin_unlock(&cur_trans
->commit_lock
);
1166 atomic_inc(&cur_trans
->use_count
);
1167 btrfs_end_transaction(trans
, root
);
1169 wait_for_commit(root
, cur_trans
);
1171 put_transaction(cur_trans
);
1176 trans
->transaction
->in_commit
= 1;
1177 trans
->transaction
->blocked
= 1;
1178 spin_unlock(&cur_trans
->commit_lock
);
1179 wake_up(&root
->fs_info
->transaction_blocked_wait
);
1181 spin_lock(&root
->fs_info
->trans_lock
);
1182 if (cur_trans
->list
.prev
!= &root
->fs_info
->trans_list
) {
1183 prev_trans
= list_entry(cur_trans
->list
.prev
,
1184 struct btrfs_transaction
, list
);
1185 if (!prev_trans
->commit_done
) {
1186 atomic_inc(&prev_trans
->use_count
);
1187 spin_unlock(&root
->fs_info
->trans_lock
);
1189 wait_for_commit(root
, prev_trans
);
1191 put_transaction(prev_trans
);
1193 spin_unlock(&root
->fs_info
->trans_lock
);
1196 spin_unlock(&root
->fs_info
->trans_lock
);
1199 if (now
< cur_trans
->start_time
|| now
- cur_trans
->start_time
< 1)
1203 int snap_pending
= 0;
1205 joined
= cur_trans
->num_joined
;
1206 if (!list_empty(&trans
->transaction
->pending_snapshots
))
1209 WARN_ON(cur_trans
!= trans
->transaction
);
1211 if (flush_on_commit
|| snap_pending
) {
1212 btrfs_start_delalloc_inodes(root
, 1);
1213 ret
= btrfs_wait_ordered_extents(root
, 0, 1);
1217 ret
= btrfs_run_delayed_items(trans
, root
);
1221 * rename don't use btrfs_join_transaction, so, once we
1222 * set the transaction to blocked above, we aren't going
1223 * to get any new ordered operations. We can safely run
1224 * it here and no for sure that nothing new will be added
1227 btrfs_run_ordered_operations(root
, 1);
1229 prepare_to_wait(&cur_trans
->writer_wait
, &wait
,
1230 TASK_UNINTERRUPTIBLE
);
1232 if (atomic_read(&cur_trans
->num_writers
) > 1)
1233 schedule_timeout(MAX_SCHEDULE_TIMEOUT
);
1234 else if (should_grow
)
1235 schedule_timeout(1);
1237 finish_wait(&cur_trans
->writer_wait
, &wait
);
1238 } while (atomic_read(&cur_trans
->num_writers
) > 1 ||
1239 (should_grow
&& cur_trans
->num_joined
!= joined
));
1242 * Ok now we need to make sure to block out any other joins while we
1243 * commit the transaction. We could have started a join before setting
1244 * no_join so make sure to wait for num_writers to == 1 again.
1246 spin_lock(&root
->fs_info
->trans_lock
);
1247 root
->fs_info
->trans_no_join
= 1;
1248 spin_unlock(&root
->fs_info
->trans_lock
);
1249 wait_event(cur_trans
->writer_wait
,
1250 atomic_read(&cur_trans
->num_writers
) == 1);
1253 * the reloc mutex makes sure that we stop
1254 * the balancing code from coming in and moving
1255 * extents around in the middle of the commit
1257 mutex_lock(&root
->fs_info
->reloc_mutex
);
1259 ret
= btrfs_run_delayed_items(trans
, root
);
1262 ret
= create_pending_snapshots(trans
, root
->fs_info
);
1265 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1269 * make sure none of the code above managed to slip in a
1272 btrfs_assert_delayed_root_empty(root
);
1274 WARN_ON(cur_trans
!= trans
->transaction
);
1276 btrfs_scrub_pause(root
);
1277 /* btrfs_commit_tree_roots is responsible for getting the
1278 * various roots consistent with each other. Every pointer
1279 * in the tree of tree roots has to point to the most up to date
1280 * root for every subvolume and other tree. So, we have to keep
1281 * the tree logging code from jumping in and changing any
1284 * At this point in the commit, there can't be any tree-log
1285 * writers, but a little lower down we drop the trans mutex
1286 * and let new people in. By holding the tree_log_mutex
1287 * from now until after the super is written, we avoid races
1288 * with the tree-log code.
1290 mutex_lock(&root
->fs_info
->tree_log_mutex
);
1292 ret
= commit_fs_roots(trans
, root
);
1295 /* commit_fs_roots gets rid of all the tree log roots, it is now
1296 * safe to free the root of tree log roots
1298 btrfs_free_log_root_tree(trans
, root
->fs_info
);
1300 ret
= commit_cowonly_roots(trans
, root
);
1303 btrfs_prepare_extent_commit(trans
, root
);
1305 cur_trans
= root
->fs_info
->running_transaction
;
1307 btrfs_set_root_node(&root
->fs_info
->tree_root
->root_item
,
1308 root
->fs_info
->tree_root
->node
);
1309 switch_commit_root(root
->fs_info
->tree_root
);
1311 btrfs_set_root_node(&root
->fs_info
->chunk_root
->root_item
,
1312 root
->fs_info
->chunk_root
->node
);
1313 switch_commit_root(root
->fs_info
->chunk_root
);
1315 update_super_roots(root
);
1317 if (!root
->fs_info
->log_root_recovering
) {
1318 btrfs_set_super_log_root(root
->fs_info
->super_copy
, 0);
1319 btrfs_set_super_log_root_level(root
->fs_info
->super_copy
, 0);
1322 memcpy(root
->fs_info
->super_for_commit
, root
->fs_info
->super_copy
,
1323 sizeof(*root
->fs_info
->super_copy
));
1325 trans
->transaction
->blocked
= 0;
1326 spin_lock(&root
->fs_info
->trans_lock
);
1327 root
->fs_info
->running_transaction
= NULL
;
1328 root
->fs_info
->trans_no_join
= 0;
1329 spin_unlock(&root
->fs_info
->trans_lock
);
1330 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1332 wake_up(&root
->fs_info
->transaction_wait
);
1334 ret
= btrfs_write_and_wait_transaction(trans
, root
);
1336 write_ctree_super(trans
, root
, 0);
1339 * the super is written, we can safely allow the tree-loggers
1340 * to go about their business
1342 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1344 btrfs_finish_extent_commit(trans
, root
);
1346 cur_trans
->commit_done
= 1;
1348 root
->fs_info
->last_trans_committed
= cur_trans
->transid
;
1350 wake_up(&cur_trans
->commit_wait
);
1352 spin_lock(&root
->fs_info
->trans_lock
);
1353 list_del_init(&cur_trans
->list
);
1354 spin_unlock(&root
->fs_info
->trans_lock
);
1356 put_transaction(cur_trans
);
1357 put_transaction(cur_trans
);
1359 trace_btrfs_transaction_commit(root
);
1361 btrfs_scrub_continue(root
);
1363 if (current
->journal_info
== trans
)
1364 current
->journal_info
= NULL
;
1366 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1368 if (current
!= root
->fs_info
->transaction_kthread
)
1369 btrfs_run_delayed_iputs(root
);
1375 * interface function to delete all the snapshots we have scheduled for deletion
1377 int btrfs_clean_old_snapshots(struct btrfs_root
*root
)
1380 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1382 spin_lock(&fs_info
->trans_lock
);
1383 list_splice_init(&fs_info
->dead_roots
, &list
);
1384 spin_unlock(&fs_info
->trans_lock
);
1386 while (!list_empty(&list
)) {
1387 root
= list_entry(list
.next
, struct btrfs_root
, root_list
);
1388 list_del(&root
->root_list
);
1390 btrfs_kill_all_delayed_nodes(root
);
1392 if (btrfs_header_backref_rev(root
->node
) <
1393 BTRFS_MIXED_BACKREF_REV
)
1394 btrfs_drop_snapshot(root
, NULL
, 0, 0);
1396 btrfs_drop_snapshot(root
, NULL
, 1, 0);