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 spin_lock_init(&cur_trans
->commit_lock
);
115 spin_lock_init(&cur_trans
->delayed_refs
.lock
);
116 INIT_LIST_HEAD(&cur_trans
->delayed_refs
.seq_head
);
118 INIT_LIST_HEAD(&cur_trans
->pending_snapshots
);
119 list_add_tail(&cur_trans
->list
, &root
->fs_info
->trans_list
);
120 extent_io_tree_init(&cur_trans
->dirty_pages
,
121 root
->fs_info
->btree_inode
->i_mapping
);
122 root
->fs_info
->generation
++;
123 cur_trans
->transid
= root
->fs_info
->generation
;
124 root
->fs_info
->running_transaction
= cur_trans
;
125 spin_unlock(&root
->fs_info
->trans_lock
);
131 * this does all the record keeping required to make sure that a reference
132 * counted root is properly recorded in a given transaction. This is required
133 * to make sure the old root from before we joined the transaction is deleted
134 * when the transaction commits
136 static int record_root_in_trans(struct btrfs_trans_handle
*trans
,
137 struct btrfs_root
*root
)
139 if (root
->ref_cows
&& root
->last_trans
< trans
->transid
) {
140 WARN_ON(root
== root
->fs_info
->extent_root
);
141 WARN_ON(root
->commit_root
!= root
->node
);
144 * see below for in_trans_setup usage rules
145 * we have the reloc mutex held now, so there
146 * is only one writer in this function
148 root
->in_trans_setup
= 1;
150 /* make sure readers find in_trans_setup before
151 * they find our root->last_trans update
155 spin_lock(&root
->fs_info
->fs_roots_radix_lock
);
156 if (root
->last_trans
== trans
->transid
) {
157 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
160 radix_tree_tag_set(&root
->fs_info
->fs_roots_radix
,
161 (unsigned long)root
->root_key
.objectid
,
162 BTRFS_ROOT_TRANS_TAG
);
163 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
164 root
->last_trans
= trans
->transid
;
166 /* this is pretty tricky. We don't want to
167 * take the relocation lock in btrfs_record_root_in_trans
168 * unless we're really doing the first setup for this root in
171 * Normally we'd use root->last_trans as a flag to decide
172 * if we want to take the expensive mutex.
174 * But, we have to set root->last_trans before we
175 * init the relocation root, otherwise, we trip over warnings
176 * in ctree.c. The solution used here is to flag ourselves
177 * with root->in_trans_setup. When this is 1, we're still
178 * fixing up the reloc trees and everyone must wait.
180 * When this is zero, they can trust root->last_trans and fly
181 * through btrfs_record_root_in_trans without having to take the
182 * lock. smp_wmb() makes sure that all the writes above are
183 * done before we pop in the zero below
185 btrfs_init_reloc_root(trans
, root
);
187 root
->in_trans_setup
= 0;
193 int btrfs_record_root_in_trans(struct btrfs_trans_handle
*trans
,
194 struct btrfs_root
*root
)
200 * see record_root_in_trans for comments about in_trans_setup usage
204 if (root
->last_trans
== trans
->transid
&&
205 !root
->in_trans_setup
)
208 mutex_lock(&root
->fs_info
->reloc_mutex
);
209 record_root_in_trans(trans
, root
);
210 mutex_unlock(&root
->fs_info
->reloc_mutex
);
215 /* wait for commit against the current transaction to become unblocked
216 * when this is done, it is safe to start a new transaction, but the current
217 * transaction might not be fully on disk.
219 static void wait_current_trans(struct btrfs_root
*root
)
221 struct btrfs_transaction
*cur_trans
;
223 spin_lock(&root
->fs_info
->trans_lock
);
224 cur_trans
= root
->fs_info
->running_transaction
;
225 if (cur_trans
&& cur_trans
->blocked
) {
226 atomic_inc(&cur_trans
->use_count
);
227 spin_unlock(&root
->fs_info
->trans_lock
);
229 wait_event(root
->fs_info
->transaction_wait
,
230 !cur_trans
->blocked
);
231 put_transaction(cur_trans
);
233 spin_unlock(&root
->fs_info
->trans_lock
);
237 enum btrfs_trans_type
{
244 static int may_wait_transaction(struct btrfs_root
*root
, int type
)
246 if (root
->fs_info
->log_root_recovering
)
249 if (type
== TRANS_USERSPACE
)
252 if (type
== TRANS_START
&&
253 !atomic_read(&root
->fs_info
->open_ioctl_trans
))
259 static struct btrfs_trans_handle
*start_transaction(struct btrfs_root
*root
,
260 u64 num_items
, int type
)
262 struct btrfs_trans_handle
*h
;
263 struct btrfs_transaction
*cur_trans
;
267 if (root
->fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
)
268 return ERR_PTR(-EROFS
);
270 if (current
->journal_info
) {
271 WARN_ON(type
!= TRANS_JOIN
&& type
!= TRANS_JOIN_NOLOCK
);
272 h
= current
->journal_info
;
274 h
->orig_rsv
= h
->block_rsv
;
280 * Do the reservation before we join the transaction so we can do all
281 * the appropriate flushing if need be.
283 if (num_items
> 0 && root
!= root
->fs_info
->chunk_root
) {
284 num_bytes
= btrfs_calc_trans_metadata_size(root
, num_items
);
285 ret
= btrfs_block_rsv_add(root
,
286 &root
->fs_info
->trans_block_rsv
,
292 h
= kmem_cache_alloc(btrfs_trans_handle_cachep
, GFP_NOFS
);
294 return ERR_PTR(-ENOMEM
);
296 if (may_wait_transaction(root
, type
))
297 wait_current_trans(root
);
300 ret
= join_transaction(root
, type
== TRANS_JOIN_NOLOCK
);
302 wait_current_trans(root
);
303 } while (ret
== -EBUSY
);
306 kmem_cache_free(btrfs_trans_handle_cachep
, h
);
310 cur_trans
= root
->fs_info
->running_transaction
;
312 h
->transid
= cur_trans
->transid
;
313 h
->transaction
= cur_trans
;
315 h
->bytes_reserved
= 0;
316 h
->delayed_ref_updates
= 0;
322 if (cur_trans
->blocked
&& may_wait_transaction(root
, type
)) {
323 btrfs_commit_transaction(h
, root
);
328 h
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
329 h
->bytes_reserved
= num_bytes
;
333 btrfs_record_root_in_trans(h
, root
);
335 if (!current
->journal_info
&& type
!= TRANS_USERSPACE
)
336 current
->journal_info
= h
;
340 struct btrfs_trans_handle
*btrfs_start_transaction(struct btrfs_root
*root
,
343 return start_transaction(root
, num_items
, TRANS_START
);
345 struct btrfs_trans_handle
*btrfs_join_transaction(struct btrfs_root
*root
)
347 return start_transaction(root
, 0, TRANS_JOIN
);
350 struct btrfs_trans_handle
*btrfs_join_transaction_nolock(struct btrfs_root
*root
)
352 return start_transaction(root
, 0, TRANS_JOIN_NOLOCK
);
355 struct btrfs_trans_handle
*btrfs_start_ioctl_transaction(struct btrfs_root
*root
)
357 return start_transaction(root
, 0, TRANS_USERSPACE
);
360 /* wait for a transaction commit to be fully complete */
361 static noinline
void wait_for_commit(struct btrfs_root
*root
,
362 struct btrfs_transaction
*commit
)
364 wait_event(commit
->commit_wait
, commit
->commit_done
);
367 int btrfs_wait_for_commit(struct btrfs_root
*root
, u64 transid
)
369 struct btrfs_transaction
*cur_trans
= NULL
, *t
;
374 if (transid
<= root
->fs_info
->last_trans_committed
)
377 /* find specified transaction */
378 spin_lock(&root
->fs_info
->trans_lock
);
379 list_for_each_entry(t
, &root
->fs_info
->trans_list
, list
) {
380 if (t
->transid
== transid
) {
382 atomic_inc(&cur_trans
->use_count
);
385 if (t
->transid
> transid
)
388 spin_unlock(&root
->fs_info
->trans_lock
);
391 goto out
; /* bad transid */
393 /* find newest transaction that is committing | committed */
394 spin_lock(&root
->fs_info
->trans_lock
);
395 list_for_each_entry_reverse(t
, &root
->fs_info
->trans_list
,
401 atomic_inc(&cur_trans
->use_count
);
405 spin_unlock(&root
->fs_info
->trans_lock
);
407 goto out
; /* nothing committing|committed */
410 wait_for_commit(root
, cur_trans
);
412 put_transaction(cur_trans
);
418 void btrfs_throttle(struct btrfs_root
*root
)
420 if (!atomic_read(&root
->fs_info
->open_ioctl_trans
))
421 wait_current_trans(root
);
424 static int should_end_transaction(struct btrfs_trans_handle
*trans
,
425 struct btrfs_root
*root
)
429 ret
= btrfs_block_rsv_check(root
, &root
->fs_info
->global_block_rsv
, 5);
433 int btrfs_should_end_transaction(struct btrfs_trans_handle
*trans
,
434 struct btrfs_root
*root
)
436 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
437 struct btrfs_block_rsv
*rsv
= trans
->block_rsv
;
441 if (cur_trans
->blocked
|| cur_trans
->delayed_refs
.flushing
)
445 * We need to do this in case we're deleting csums so the global block
446 * rsv get's used instead of the csum block rsv.
448 trans
->block_rsv
= NULL
;
450 updates
= trans
->delayed_ref_updates
;
451 trans
->delayed_ref_updates
= 0;
453 btrfs_run_delayed_refs(trans
, root
, updates
);
455 trans
->block_rsv
= rsv
;
457 return should_end_transaction(trans
, root
);
460 static int __btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
461 struct btrfs_root
*root
, int throttle
, int lock
)
463 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
464 struct btrfs_fs_info
*info
= root
->fs_info
;
467 if (--trans
->use_count
) {
468 trans
->block_rsv
= trans
->orig_rsv
;
472 btrfs_trans_release_metadata(trans
, root
);
473 trans
->block_rsv
= NULL
;
475 unsigned long cur
= trans
->delayed_ref_updates
;
476 trans
->delayed_ref_updates
= 0;
478 trans
->transaction
->delayed_refs
.num_heads_ready
> 64) {
479 trans
->delayed_ref_updates
= 0;
482 * do a full flush if the transaction is trying
485 if (trans
->transaction
->delayed_refs
.flushing
)
487 btrfs_run_delayed_refs(trans
, root
, cur
);
494 if (lock
&& !atomic_read(&root
->fs_info
->open_ioctl_trans
) &&
495 should_end_transaction(trans
, root
)) {
496 trans
->transaction
->blocked
= 1;
500 if (lock
&& cur_trans
->blocked
&& !cur_trans
->in_commit
) {
503 * We may race with somebody else here so end up having
504 * to call end_transaction on ourselves again, so inc
508 return btrfs_commit_transaction(trans
, root
);
510 wake_up_process(info
->transaction_kthread
);
514 WARN_ON(cur_trans
!= info
->running_transaction
);
515 WARN_ON(atomic_read(&cur_trans
->num_writers
) < 1);
516 atomic_dec(&cur_trans
->num_writers
);
519 if (waitqueue_active(&cur_trans
->writer_wait
))
520 wake_up(&cur_trans
->writer_wait
);
521 put_transaction(cur_trans
);
523 if (current
->journal_info
== trans
)
524 current
->journal_info
= NULL
;
525 memset(trans
, 0, sizeof(*trans
));
526 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
529 btrfs_run_delayed_iputs(root
);
534 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
535 struct btrfs_root
*root
)
539 ret
= __btrfs_end_transaction(trans
, root
, 0, 1);
545 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
,
546 struct btrfs_root
*root
)
550 ret
= __btrfs_end_transaction(trans
, root
, 1, 1);
556 int btrfs_end_transaction_nolock(struct btrfs_trans_handle
*trans
,
557 struct btrfs_root
*root
)
561 ret
= __btrfs_end_transaction(trans
, root
, 0, 0);
567 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle
*trans
,
568 struct btrfs_root
*root
)
570 return __btrfs_end_transaction(trans
, root
, 1, 1);
574 * when btree blocks are allocated, they have some corresponding bits set for
575 * them in one of two extent_io trees. This is used to make sure all of
576 * those extents are sent to disk but does not wait on them
578 int btrfs_write_marked_extents(struct btrfs_root
*root
,
579 struct extent_io_tree
*dirty_pages
, int mark
)
583 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
587 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
589 convert_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
, mark
,
591 err
= filemap_fdatawrite_range(mapping
, start
, end
);
603 * when btree blocks are allocated, they have some corresponding bits set for
604 * them in one of two extent_io trees. This is used to make sure all of
605 * those extents are on disk for transaction or log commit. We wait
606 * on all the pages and clear them from the dirty pages state tree
608 int btrfs_wait_marked_extents(struct btrfs_root
*root
,
609 struct extent_io_tree
*dirty_pages
, int mark
)
613 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
617 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
619 clear_extent_bits(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
, GFP_NOFS
);
620 err
= filemap_fdatawait_range(mapping
, start
, end
);
632 * when btree blocks are allocated, they have some corresponding bits set for
633 * them in one of two extent_io trees. This is used to make sure all of
634 * those extents are on disk for transaction or log commit
636 int btrfs_write_and_wait_marked_extents(struct btrfs_root
*root
,
637 struct extent_io_tree
*dirty_pages
, int mark
)
642 ret
= btrfs_write_marked_extents(root
, dirty_pages
, mark
);
643 ret2
= btrfs_wait_marked_extents(root
, dirty_pages
, mark
);
652 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
,
653 struct btrfs_root
*root
)
655 if (!trans
|| !trans
->transaction
) {
656 struct inode
*btree_inode
;
657 btree_inode
= root
->fs_info
->btree_inode
;
658 return filemap_write_and_wait(btree_inode
->i_mapping
);
660 return btrfs_write_and_wait_marked_extents(root
,
661 &trans
->transaction
->dirty_pages
,
666 * this is used to update the root pointer in the tree of tree roots.
668 * But, in the case of the extent allocation tree, updating the root
669 * pointer may allocate blocks which may change the root of the extent
672 * So, this loops and repeats and makes sure the cowonly root didn't
673 * change while the root pointer was being updated in the metadata.
675 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
676 struct btrfs_root
*root
)
681 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
683 old_root_used
= btrfs_root_used(&root
->root_item
);
684 btrfs_write_dirty_block_groups(trans
, root
);
687 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
688 if (old_root_bytenr
== root
->node
->start
&&
689 old_root_used
== btrfs_root_used(&root
->root_item
))
692 btrfs_set_root_node(&root
->root_item
, root
->node
);
693 ret
= btrfs_update_root(trans
, tree_root
,
698 old_root_used
= btrfs_root_used(&root
->root_item
);
699 ret
= btrfs_write_dirty_block_groups(trans
, root
);
703 if (root
!= root
->fs_info
->extent_root
)
704 switch_commit_root(root
);
710 * update all the cowonly tree roots on disk
712 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
,
713 struct btrfs_root
*root
)
715 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
716 struct list_head
*next
;
717 struct extent_buffer
*eb
;
720 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
723 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
724 btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
, 0, &eb
);
725 btrfs_tree_unlock(eb
);
726 free_extent_buffer(eb
);
728 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
731 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
732 next
= fs_info
->dirty_cowonly_roots
.next
;
734 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
736 update_cowonly_root(trans
, root
);
739 down_write(&fs_info
->extent_commit_sem
);
740 switch_commit_root(fs_info
->extent_root
);
741 up_write(&fs_info
->extent_commit_sem
);
747 * dead roots are old snapshots that need to be deleted. This allocates
748 * a dirty root struct and adds it into the list of dead roots that need to
751 int btrfs_add_dead_root(struct btrfs_root
*root
)
753 spin_lock(&root
->fs_info
->trans_lock
);
754 list_add(&root
->root_list
, &root
->fs_info
->dead_roots
);
755 spin_unlock(&root
->fs_info
->trans_lock
);
760 * update all the cowonly tree roots on disk
762 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
,
763 struct btrfs_root
*root
)
765 struct btrfs_root
*gang
[8];
766 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
771 spin_lock(&fs_info
->fs_roots_radix_lock
);
773 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
776 BTRFS_ROOT_TRANS_TAG
);
779 for (i
= 0; i
< ret
; i
++) {
781 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
782 (unsigned long)root
->root_key
.objectid
,
783 BTRFS_ROOT_TRANS_TAG
);
784 spin_unlock(&fs_info
->fs_roots_radix_lock
);
786 btrfs_free_log(trans
, root
);
787 btrfs_update_reloc_root(trans
, root
);
788 btrfs_orphan_commit_root(trans
, root
);
790 btrfs_save_ino_cache(root
, trans
);
792 /* see comments in should_cow_block() */
796 if (root
->commit_root
!= root
->node
) {
797 mutex_lock(&root
->fs_commit_mutex
);
798 switch_commit_root(root
);
799 btrfs_unpin_free_ino(root
);
800 mutex_unlock(&root
->fs_commit_mutex
);
802 btrfs_set_root_node(&root
->root_item
,
806 err
= btrfs_update_root(trans
, fs_info
->tree_root
,
809 spin_lock(&fs_info
->fs_roots_radix_lock
);
814 spin_unlock(&fs_info
->fs_roots_radix_lock
);
819 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
820 * otherwise every leaf in the btree is read and defragged.
822 int btrfs_defrag_root(struct btrfs_root
*root
, int cacheonly
)
824 struct btrfs_fs_info
*info
= root
->fs_info
;
825 struct btrfs_trans_handle
*trans
;
829 if (xchg(&root
->defrag_running
, 1))
833 trans
= btrfs_start_transaction(root
, 0);
835 return PTR_ERR(trans
);
837 ret
= btrfs_defrag_leaves(trans
, root
, cacheonly
);
839 nr
= trans
->blocks_used
;
840 btrfs_end_transaction(trans
, root
);
841 btrfs_btree_balance_dirty(info
->tree_root
, nr
);
844 if (btrfs_fs_closing(root
->fs_info
) || ret
!= -EAGAIN
)
847 root
->defrag_running
= 0;
852 * new snapshots need to be created at a very specific time in the
853 * transaction commit. This does the actual creation
855 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
856 struct btrfs_fs_info
*fs_info
,
857 struct btrfs_pending_snapshot
*pending
)
859 struct btrfs_key key
;
860 struct btrfs_root_item
*new_root_item
;
861 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
862 struct btrfs_root
*root
= pending
->root
;
863 struct btrfs_root
*parent_root
;
864 struct btrfs_block_rsv
*rsv
;
865 struct inode
*parent_inode
;
866 struct dentry
*parent
;
867 struct dentry
*dentry
;
868 struct extent_buffer
*tmp
;
869 struct extent_buffer
*old
;
876 rsv
= trans
->block_rsv
;
878 new_root_item
= kmalloc(sizeof(*new_root_item
), GFP_NOFS
);
879 if (!new_root_item
) {
880 pending
->error
= -ENOMEM
;
884 ret
= btrfs_find_free_objectid(tree_root
, &objectid
);
886 pending
->error
= ret
;
890 btrfs_reloc_pre_snapshot(trans
, pending
, &to_reserve
);
892 if (to_reserve
> 0) {
893 ret
= btrfs_block_rsv_add_noflush(root
, &pending
->block_rsv
,
896 pending
->error
= ret
;
901 key
.objectid
= objectid
;
902 key
.offset
= (u64
)-1;
903 key
.type
= BTRFS_ROOT_ITEM_KEY
;
905 trans
->block_rsv
= &pending
->block_rsv
;
907 dentry
= pending
->dentry
;
908 parent
= dget_parent(dentry
);
909 parent_inode
= parent
->d_inode
;
910 parent_root
= BTRFS_I(parent_inode
)->root
;
911 record_root_in_trans(trans
, parent_root
);
914 * insert the directory item
916 ret
= btrfs_set_inode_index(parent_inode
, &index
);
918 ret
= btrfs_insert_dir_item(trans
, parent_root
,
919 dentry
->d_name
.name
, dentry
->d_name
.len
,
921 BTRFS_FT_DIR
, index
);
924 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
925 dentry
->d_name
.len
* 2);
926 ret
= btrfs_update_inode(trans
, parent_root
, parent_inode
);
930 * pull in the delayed directory update
931 * and the delayed inode item
932 * otherwise we corrupt the FS during
935 ret
= btrfs_run_delayed_items(trans
, root
);
938 record_root_in_trans(trans
, root
);
939 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
940 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
941 btrfs_check_and_init_root_item(new_root_item
);
943 root_flags
= btrfs_root_flags(new_root_item
);
944 if (pending
->readonly
)
945 root_flags
|= BTRFS_ROOT_SUBVOL_RDONLY
;
947 root_flags
&= ~BTRFS_ROOT_SUBVOL_RDONLY
;
948 btrfs_set_root_flags(new_root_item
, root_flags
);
950 old
= btrfs_lock_root_node(root
);
951 btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
);
952 btrfs_set_lock_blocking(old
);
954 btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
955 btrfs_tree_unlock(old
);
956 free_extent_buffer(old
);
958 /* see comments in should_cow_block() */
962 btrfs_set_root_node(new_root_item
, tmp
);
963 /* record when the snapshot was created in key.offset */
964 key
.offset
= trans
->transid
;
965 ret
= btrfs_insert_root(trans
, tree_root
, &key
, new_root_item
);
966 btrfs_tree_unlock(tmp
);
967 free_extent_buffer(tmp
);
971 * insert root back/forward references
973 ret
= btrfs_add_root_ref(trans
, tree_root
, objectid
,
974 parent_root
->root_key
.objectid
,
975 btrfs_ino(parent_inode
), index
,
976 dentry
->d_name
.name
, dentry
->d_name
.len
);
980 key
.offset
= (u64
)-1;
981 pending
->snap
= btrfs_read_fs_root_no_name(root
->fs_info
, &key
);
982 BUG_ON(IS_ERR(pending
->snap
));
984 btrfs_reloc_post_snapshot(trans
, pending
);
986 kfree(new_root_item
);
987 trans
->block_rsv
= rsv
;
988 btrfs_block_rsv_release(root
, &pending
->block_rsv
, (u64
)-1);
993 * create all the snapshots we've scheduled for creation
995 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
,
996 struct btrfs_fs_info
*fs_info
)
998 struct btrfs_pending_snapshot
*pending
;
999 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
1002 list_for_each_entry(pending
, head
, list
) {
1003 ret
= create_pending_snapshot(trans
, fs_info
, pending
);
1009 static void update_super_roots(struct btrfs_root
*root
)
1011 struct btrfs_root_item
*root_item
;
1012 struct btrfs_super_block
*super
;
1014 super
= root
->fs_info
->super_copy
;
1016 root_item
= &root
->fs_info
->chunk_root
->root_item
;
1017 super
->chunk_root
= root_item
->bytenr
;
1018 super
->chunk_root_generation
= root_item
->generation
;
1019 super
->chunk_root_level
= root_item
->level
;
1021 root_item
= &root
->fs_info
->tree_root
->root_item
;
1022 super
->root
= root_item
->bytenr
;
1023 super
->generation
= root_item
->generation
;
1024 super
->root_level
= root_item
->level
;
1025 if (btrfs_test_opt(root
, SPACE_CACHE
))
1026 super
->cache_generation
= root_item
->generation
;
1029 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
1032 spin_lock(&info
->trans_lock
);
1033 if (info
->running_transaction
)
1034 ret
= info
->running_transaction
->in_commit
;
1035 spin_unlock(&info
->trans_lock
);
1039 int btrfs_transaction_blocked(struct btrfs_fs_info
*info
)
1042 spin_lock(&info
->trans_lock
);
1043 if (info
->running_transaction
)
1044 ret
= info
->running_transaction
->blocked
;
1045 spin_unlock(&info
->trans_lock
);
1050 * wait for the current transaction commit to start and block subsequent
1053 static void wait_current_trans_commit_start(struct btrfs_root
*root
,
1054 struct btrfs_transaction
*trans
)
1056 wait_event(root
->fs_info
->transaction_blocked_wait
, trans
->in_commit
);
1060 * wait for the current transaction to start and then become unblocked.
1063 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root
*root
,
1064 struct btrfs_transaction
*trans
)
1066 wait_event(root
->fs_info
->transaction_wait
,
1067 trans
->commit_done
|| (trans
->in_commit
&& !trans
->blocked
));
1071 * commit transactions asynchronously. once btrfs_commit_transaction_async
1072 * returns, any subsequent transaction will not be allowed to join.
1074 struct btrfs_async_commit
{
1075 struct btrfs_trans_handle
*newtrans
;
1076 struct btrfs_root
*root
;
1077 struct delayed_work work
;
1080 static void do_async_commit(struct work_struct
*work
)
1082 struct btrfs_async_commit
*ac
=
1083 container_of(work
, struct btrfs_async_commit
, work
.work
);
1085 btrfs_commit_transaction(ac
->newtrans
, ac
->root
);
1089 int btrfs_commit_transaction_async(struct btrfs_trans_handle
*trans
,
1090 struct btrfs_root
*root
,
1091 int wait_for_unblock
)
1093 struct btrfs_async_commit
*ac
;
1094 struct btrfs_transaction
*cur_trans
;
1096 ac
= kmalloc(sizeof(*ac
), GFP_NOFS
);
1100 INIT_DELAYED_WORK(&ac
->work
, do_async_commit
);
1102 ac
->newtrans
= btrfs_join_transaction(root
);
1103 if (IS_ERR(ac
->newtrans
)) {
1104 int err
= PTR_ERR(ac
->newtrans
);
1109 /* take transaction reference */
1110 cur_trans
= trans
->transaction
;
1111 atomic_inc(&cur_trans
->use_count
);
1113 btrfs_end_transaction(trans
, root
);
1114 schedule_delayed_work(&ac
->work
, 0);
1116 /* wait for transaction to start and unblock */
1117 if (wait_for_unblock
)
1118 wait_current_trans_commit_start_and_unblock(root
, cur_trans
);
1120 wait_current_trans_commit_start(root
, cur_trans
);
1122 if (current
->journal_info
== trans
)
1123 current
->journal_info
= NULL
;
1125 put_transaction(cur_trans
);
1130 * btrfs_transaction state sequence:
1131 * in_commit = 0, blocked = 0 (initial)
1132 * in_commit = 1, blocked = 1
1136 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
,
1137 struct btrfs_root
*root
)
1139 unsigned long joined
= 0;
1140 struct btrfs_transaction
*cur_trans
;
1141 struct btrfs_transaction
*prev_trans
= NULL
;
1144 int should_grow
= 0;
1145 unsigned long now
= get_seconds();
1146 int flush_on_commit
= btrfs_test_opt(root
, FLUSHONCOMMIT
);
1148 btrfs_run_ordered_operations(root
, 0);
1150 btrfs_trans_release_metadata(trans
, root
);
1151 trans
->block_rsv
= NULL
;
1153 /* make a pass through all the delayed refs we have so far
1154 * any runnings procs may add more while we are here
1156 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1159 cur_trans
= trans
->transaction
;
1161 * set the flushing flag so procs in this transaction have to
1162 * start sending their work down.
1164 cur_trans
->delayed_refs
.flushing
= 1;
1166 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1169 spin_lock(&cur_trans
->commit_lock
);
1170 if (cur_trans
->in_commit
) {
1171 spin_unlock(&cur_trans
->commit_lock
);
1172 atomic_inc(&cur_trans
->use_count
);
1173 btrfs_end_transaction(trans
, root
);
1175 wait_for_commit(root
, cur_trans
);
1177 put_transaction(cur_trans
);
1182 trans
->transaction
->in_commit
= 1;
1183 trans
->transaction
->blocked
= 1;
1184 spin_unlock(&cur_trans
->commit_lock
);
1185 wake_up(&root
->fs_info
->transaction_blocked_wait
);
1187 spin_lock(&root
->fs_info
->trans_lock
);
1188 if (cur_trans
->list
.prev
!= &root
->fs_info
->trans_list
) {
1189 prev_trans
= list_entry(cur_trans
->list
.prev
,
1190 struct btrfs_transaction
, list
);
1191 if (!prev_trans
->commit_done
) {
1192 atomic_inc(&prev_trans
->use_count
);
1193 spin_unlock(&root
->fs_info
->trans_lock
);
1195 wait_for_commit(root
, prev_trans
);
1197 put_transaction(prev_trans
);
1199 spin_unlock(&root
->fs_info
->trans_lock
);
1202 spin_unlock(&root
->fs_info
->trans_lock
);
1205 if (now
< cur_trans
->start_time
|| now
- cur_trans
->start_time
< 1)
1209 int snap_pending
= 0;
1211 joined
= cur_trans
->num_joined
;
1212 if (!list_empty(&trans
->transaction
->pending_snapshots
))
1215 WARN_ON(cur_trans
!= trans
->transaction
);
1217 if (flush_on_commit
|| snap_pending
) {
1218 btrfs_start_delalloc_inodes(root
, 1);
1219 ret
= btrfs_wait_ordered_extents(root
, 0, 1);
1223 ret
= btrfs_run_delayed_items(trans
, root
);
1227 * rename don't use btrfs_join_transaction, so, once we
1228 * set the transaction to blocked above, we aren't going
1229 * to get any new ordered operations. We can safely run
1230 * it here and no for sure that nothing new will be added
1233 btrfs_run_ordered_operations(root
, 1);
1235 prepare_to_wait(&cur_trans
->writer_wait
, &wait
,
1236 TASK_UNINTERRUPTIBLE
);
1238 if (atomic_read(&cur_trans
->num_writers
) > 1)
1239 schedule_timeout(MAX_SCHEDULE_TIMEOUT
);
1240 else if (should_grow
)
1241 schedule_timeout(1);
1243 finish_wait(&cur_trans
->writer_wait
, &wait
);
1244 } while (atomic_read(&cur_trans
->num_writers
) > 1 ||
1245 (should_grow
&& cur_trans
->num_joined
!= joined
));
1248 * Ok now we need to make sure to block out any other joins while we
1249 * commit the transaction. We could have started a join before setting
1250 * no_join so make sure to wait for num_writers to == 1 again.
1252 spin_lock(&root
->fs_info
->trans_lock
);
1253 root
->fs_info
->trans_no_join
= 1;
1254 spin_unlock(&root
->fs_info
->trans_lock
);
1255 wait_event(cur_trans
->writer_wait
,
1256 atomic_read(&cur_trans
->num_writers
) == 1);
1259 * the reloc mutex makes sure that we stop
1260 * the balancing code from coming in and moving
1261 * extents around in the middle of the commit
1263 mutex_lock(&root
->fs_info
->reloc_mutex
);
1265 ret
= btrfs_run_delayed_items(trans
, root
);
1268 ret
= create_pending_snapshots(trans
, root
->fs_info
);
1271 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1275 * make sure none of the code above managed to slip in a
1278 btrfs_assert_delayed_root_empty(root
);
1280 WARN_ON(cur_trans
!= trans
->transaction
);
1282 btrfs_scrub_pause(root
);
1283 /* btrfs_commit_tree_roots is responsible for getting the
1284 * various roots consistent with each other. Every pointer
1285 * in the tree of tree roots has to point to the most up to date
1286 * root for every subvolume and other tree. So, we have to keep
1287 * the tree logging code from jumping in and changing any
1290 * At this point in the commit, there can't be any tree-log
1291 * writers, but a little lower down we drop the trans mutex
1292 * and let new people in. By holding the tree_log_mutex
1293 * from now until after the super is written, we avoid races
1294 * with the tree-log code.
1296 mutex_lock(&root
->fs_info
->tree_log_mutex
);
1298 ret
= commit_fs_roots(trans
, root
);
1301 /* commit_fs_roots gets rid of all the tree log roots, it is now
1302 * safe to free the root of tree log roots
1304 btrfs_free_log_root_tree(trans
, root
->fs_info
);
1306 ret
= commit_cowonly_roots(trans
, root
);
1309 btrfs_prepare_extent_commit(trans
, root
);
1311 cur_trans
= root
->fs_info
->running_transaction
;
1313 btrfs_set_root_node(&root
->fs_info
->tree_root
->root_item
,
1314 root
->fs_info
->tree_root
->node
);
1315 switch_commit_root(root
->fs_info
->tree_root
);
1317 btrfs_set_root_node(&root
->fs_info
->chunk_root
->root_item
,
1318 root
->fs_info
->chunk_root
->node
);
1319 switch_commit_root(root
->fs_info
->chunk_root
);
1321 update_super_roots(root
);
1323 if (!root
->fs_info
->log_root_recovering
) {
1324 btrfs_set_super_log_root(root
->fs_info
->super_copy
, 0);
1325 btrfs_set_super_log_root_level(root
->fs_info
->super_copy
, 0);
1328 memcpy(root
->fs_info
->super_for_commit
, root
->fs_info
->super_copy
,
1329 sizeof(*root
->fs_info
->super_copy
));
1331 trans
->transaction
->blocked
= 0;
1332 spin_lock(&root
->fs_info
->trans_lock
);
1333 root
->fs_info
->running_transaction
= NULL
;
1334 root
->fs_info
->trans_no_join
= 0;
1335 spin_unlock(&root
->fs_info
->trans_lock
);
1336 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1338 wake_up(&root
->fs_info
->transaction_wait
);
1340 ret
= btrfs_write_and_wait_transaction(trans
, root
);
1342 write_ctree_super(trans
, root
, 0);
1345 * the super is written, we can safely allow the tree-loggers
1346 * to go about their business
1348 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1350 btrfs_finish_extent_commit(trans
, root
);
1352 cur_trans
->commit_done
= 1;
1354 root
->fs_info
->last_trans_committed
= cur_trans
->transid
;
1356 wake_up(&cur_trans
->commit_wait
);
1358 spin_lock(&root
->fs_info
->trans_lock
);
1359 list_del_init(&cur_trans
->list
);
1360 spin_unlock(&root
->fs_info
->trans_lock
);
1362 put_transaction(cur_trans
);
1363 put_transaction(cur_trans
);
1365 trace_btrfs_transaction_commit(root
);
1367 btrfs_scrub_continue(root
);
1369 if (current
->journal_info
== trans
)
1370 current
->journal_info
= NULL
;
1372 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1374 if (current
!= root
->fs_info
->transaction_kthread
)
1375 btrfs_run_delayed_iputs(root
);
1381 * interface function to delete all the snapshots we have scheduled for deletion
1383 int btrfs_clean_old_snapshots(struct btrfs_root
*root
)
1386 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1388 spin_lock(&fs_info
->trans_lock
);
1389 list_splice_init(&fs_info
->dead_roots
, &list
);
1390 spin_unlock(&fs_info
->trans_lock
);
1392 while (!list_empty(&list
)) {
1393 root
= list_entry(list
.next
, struct btrfs_root
, root_list
);
1394 list_del(&root
->root_list
);
1396 btrfs_kill_all_delayed_nodes(root
);
1398 if (btrfs_header_backref_rev(root
->node
) <
1399 BTRFS_MIXED_BACKREF_REV
)
1400 btrfs_drop_snapshot(root
, NULL
, 0, 0);
1402 btrfs_drop_snapshot(root
, NULL
, 1, 0);