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 memset(transaction
, 0, sizeof(*transaction
));
40 kmem_cache_free(btrfs_transaction_cachep
, transaction
);
44 static noinline
void switch_commit_root(struct btrfs_root
*root
)
46 free_extent_buffer(root
->commit_root
);
47 root
->commit_root
= btrfs_root_node(root
);
51 * either allocate a new transaction or hop into the existing one
53 static noinline
int join_transaction(struct btrfs_root
*root
, int nofail
)
55 struct btrfs_transaction
*cur_trans
;
57 spin_lock(&root
->fs_info
->trans_lock
);
59 if (root
->fs_info
->trans_no_join
) {
61 spin_unlock(&root
->fs_info
->trans_lock
);
66 cur_trans
= root
->fs_info
->running_transaction
;
68 atomic_inc(&cur_trans
->use_count
);
69 atomic_inc(&cur_trans
->num_writers
);
70 cur_trans
->num_joined
++;
71 spin_unlock(&root
->fs_info
->trans_lock
);
74 spin_unlock(&root
->fs_info
->trans_lock
);
76 cur_trans
= kmem_cache_alloc(btrfs_transaction_cachep
, GFP_NOFS
);
80 spin_lock(&root
->fs_info
->trans_lock
);
81 if (root
->fs_info
->running_transaction
) {
83 * someone started a transaction after we unlocked. Make sure
84 * to redo the trans_no_join checks above
86 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
87 cur_trans
= root
->fs_info
->running_transaction
;
91 atomic_set(&cur_trans
->num_writers
, 1);
92 cur_trans
->num_joined
= 0;
93 init_waitqueue_head(&cur_trans
->writer_wait
);
94 init_waitqueue_head(&cur_trans
->commit_wait
);
95 cur_trans
->in_commit
= 0;
96 cur_trans
->blocked
= 0;
98 * One for this trans handle, one so it will live on until we
99 * commit the transaction.
101 atomic_set(&cur_trans
->use_count
, 2);
102 cur_trans
->commit_done
= 0;
103 cur_trans
->start_time
= get_seconds();
105 cur_trans
->delayed_refs
.root
= RB_ROOT
;
106 cur_trans
->delayed_refs
.num_entries
= 0;
107 cur_trans
->delayed_refs
.num_heads_ready
= 0;
108 cur_trans
->delayed_refs
.num_heads
= 0;
109 cur_trans
->delayed_refs
.flushing
= 0;
110 cur_trans
->delayed_refs
.run_delayed_start
= 0;
111 spin_lock_init(&cur_trans
->commit_lock
);
112 spin_lock_init(&cur_trans
->delayed_refs
.lock
);
114 INIT_LIST_HEAD(&cur_trans
->pending_snapshots
);
115 list_add_tail(&cur_trans
->list
, &root
->fs_info
->trans_list
);
116 extent_io_tree_init(&cur_trans
->dirty_pages
,
117 root
->fs_info
->btree_inode
->i_mapping
);
118 root
->fs_info
->generation
++;
119 cur_trans
->transid
= root
->fs_info
->generation
;
120 root
->fs_info
->running_transaction
= cur_trans
;
121 spin_unlock(&root
->fs_info
->trans_lock
);
127 * this does all the record keeping required to make sure that a reference
128 * counted root is properly recorded in a given transaction. This is required
129 * to make sure the old root from before we joined the transaction is deleted
130 * when the transaction commits
132 static int record_root_in_trans(struct btrfs_trans_handle
*trans
,
133 struct btrfs_root
*root
)
135 if (root
->ref_cows
&& root
->last_trans
< trans
->transid
) {
136 WARN_ON(root
== root
->fs_info
->extent_root
);
137 WARN_ON(root
->commit_root
!= root
->node
);
140 * see below for in_trans_setup usage rules
141 * we have the reloc mutex held now, so there
142 * is only one writer in this function
144 root
->in_trans_setup
= 1;
146 /* make sure readers find in_trans_setup before
147 * they find our root->last_trans update
151 spin_lock(&root
->fs_info
->fs_roots_radix_lock
);
152 if (root
->last_trans
== trans
->transid
) {
153 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
156 radix_tree_tag_set(&root
->fs_info
->fs_roots_radix
,
157 (unsigned long)root
->root_key
.objectid
,
158 BTRFS_ROOT_TRANS_TAG
);
159 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
160 root
->last_trans
= trans
->transid
;
162 /* this is pretty tricky. We don't want to
163 * take the relocation lock in btrfs_record_root_in_trans
164 * unless we're really doing the first setup for this root in
167 * Normally we'd use root->last_trans as a flag to decide
168 * if we want to take the expensive mutex.
170 * But, we have to set root->last_trans before we
171 * init the relocation root, otherwise, we trip over warnings
172 * in ctree.c. The solution used here is to flag ourselves
173 * with root->in_trans_setup. When this is 1, we're still
174 * fixing up the reloc trees and everyone must wait.
176 * When this is zero, they can trust root->last_trans and fly
177 * through btrfs_record_root_in_trans without having to take the
178 * lock. smp_wmb() makes sure that all the writes above are
179 * done before we pop in the zero below
181 btrfs_init_reloc_root(trans
, root
);
183 root
->in_trans_setup
= 0;
189 int btrfs_record_root_in_trans(struct btrfs_trans_handle
*trans
,
190 struct btrfs_root
*root
)
196 * see record_root_in_trans for comments about in_trans_setup usage
200 if (root
->last_trans
== trans
->transid
&&
201 !root
->in_trans_setup
)
204 mutex_lock(&root
->fs_info
->reloc_mutex
);
205 record_root_in_trans(trans
, root
);
206 mutex_unlock(&root
->fs_info
->reloc_mutex
);
211 /* wait for commit against the current transaction to become unblocked
212 * when this is done, it is safe to start a new transaction, but the current
213 * transaction might not be fully on disk.
215 static void wait_current_trans(struct btrfs_root
*root
)
217 struct btrfs_transaction
*cur_trans
;
219 spin_lock(&root
->fs_info
->trans_lock
);
220 cur_trans
= root
->fs_info
->running_transaction
;
221 if (cur_trans
&& cur_trans
->blocked
) {
222 atomic_inc(&cur_trans
->use_count
);
223 spin_unlock(&root
->fs_info
->trans_lock
);
225 wait_event(root
->fs_info
->transaction_wait
,
226 !cur_trans
->blocked
);
227 put_transaction(cur_trans
);
229 spin_unlock(&root
->fs_info
->trans_lock
);
233 enum btrfs_trans_type
{
240 static int may_wait_transaction(struct btrfs_root
*root
, int type
)
242 if (root
->fs_info
->log_root_recovering
)
245 if (type
== TRANS_USERSPACE
)
248 if (type
== TRANS_START
&&
249 !atomic_read(&root
->fs_info
->open_ioctl_trans
))
255 static struct btrfs_trans_handle
*start_transaction(struct btrfs_root
*root
,
256 u64 num_items
, int type
)
258 struct btrfs_trans_handle
*h
;
259 struct btrfs_transaction
*cur_trans
;
263 if (root
->fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
)
264 return ERR_PTR(-EROFS
);
266 if (current
->journal_info
) {
267 WARN_ON(type
!= TRANS_JOIN
&& type
!= TRANS_JOIN_NOLOCK
);
268 h
= current
->journal_info
;
270 h
->orig_rsv
= h
->block_rsv
;
276 * Do the reservation before we join the transaction so we can do all
277 * the appropriate flushing if need be.
279 if (num_items
> 0 && root
!= root
->fs_info
->chunk_root
) {
280 num_bytes
= btrfs_calc_trans_metadata_size(root
, num_items
);
281 ret
= btrfs_block_rsv_add(root
,
282 &root
->fs_info
->trans_block_rsv
,
288 h
= kmem_cache_alloc(btrfs_trans_handle_cachep
, GFP_NOFS
);
290 return ERR_PTR(-ENOMEM
);
292 if (may_wait_transaction(root
, type
))
293 wait_current_trans(root
);
296 ret
= join_transaction(root
, type
== TRANS_JOIN_NOLOCK
);
298 wait_current_trans(root
);
299 } while (ret
== -EBUSY
);
302 kmem_cache_free(btrfs_trans_handle_cachep
, h
);
306 cur_trans
= root
->fs_info
->running_transaction
;
308 h
->transid
= cur_trans
->transid
;
309 h
->transaction
= cur_trans
;
311 h
->bytes_reserved
= 0;
312 h
->delayed_ref_updates
= 0;
318 if (cur_trans
->blocked
&& may_wait_transaction(root
, type
)) {
319 btrfs_commit_transaction(h
, root
);
324 h
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
325 h
->bytes_reserved
= num_bytes
;
329 btrfs_record_root_in_trans(h
, root
);
331 if (!current
->journal_info
&& type
!= TRANS_USERSPACE
)
332 current
->journal_info
= h
;
336 struct btrfs_trans_handle
*btrfs_start_transaction(struct btrfs_root
*root
,
339 return start_transaction(root
, num_items
, TRANS_START
);
341 struct btrfs_trans_handle
*btrfs_join_transaction(struct btrfs_root
*root
)
343 return start_transaction(root
, 0, TRANS_JOIN
);
346 struct btrfs_trans_handle
*btrfs_join_transaction_nolock(struct btrfs_root
*root
)
348 return start_transaction(root
, 0, TRANS_JOIN_NOLOCK
);
351 struct btrfs_trans_handle
*btrfs_start_ioctl_transaction(struct btrfs_root
*root
)
353 return start_transaction(root
, 0, TRANS_USERSPACE
);
356 /* wait for a transaction commit to be fully complete */
357 static noinline
void wait_for_commit(struct btrfs_root
*root
,
358 struct btrfs_transaction
*commit
)
360 wait_event(commit
->commit_wait
, commit
->commit_done
);
363 int btrfs_wait_for_commit(struct btrfs_root
*root
, u64 transid
)
365 struct btrfs_transaction
*cur_trans
= NULL
, *t
;
370 if (transid
<= root
->fs_info
->last_trans_committed
)
373 /* find specified transaction */
374 spin_lock(&root
->fs_info
->trans_lock
);
375 list_for_each_entry(t
, &root
->fs_info
->trans_list
, list
) {
376 if (t
->transid
== transid
) {
378 atomic_inc(&cur_trans
->use_count
);
381 if (t
->transid
> transid
)
384 spin_unlock(&root
->fs_info
->trans_lock
);
387 goto out
; /* bad transid */
389 /* find newest transaction that is committing | committed */
390 spin_lock(&root
->fs_info
->trans_lock
);
391 list_for_each_entry_reverse(t
, &root
->fs_info
->trans_list
,
397 atomic_inc(&cur_trans
->use_count
);
401 spin_unlock(&root
->fs_info
->trans_lock
);
403 goto out
; /* nothing committing|committed */
406 wait_for_commit(root
, cur_trans
);
408 put_transaction(cur_trans
);
414 void btrfs_throttle(struct btrfs_root
*root
)
416 if (!atomic_read(&root
->fs_info
->open_ioctl_trans
))
417 wait_current_trans(root
);
420 static int should_end_transaction(struct btrfs_trans_handle
*trans
,
421 struct btrfs_root
*root
)
425 ret
= btrfs_block_rsv_check(root
, &root
->fs_info
->global_block_rsv
, 5);
429 int btrfs_should_end_transaction(struct btrfs_trans_handle
*trans
,
430 struct btrfs_root
*root
)
432 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
433 struct btrfs_block_rsv
*rsv
= trans
->block_rsv
;
437 if (cur_trans
->blocked
|| cur_trans
->delayed_refs
.flushing
)
441 * We need to do this in case we're deleting csums so the global block
442 * rsv get's used instead of the csum block rsv.
444 trans
->block_rsv
= NULL
;
446 updates
= trans
->delayed_ref_updates
;
447 trans
->delayed_ref_updates
= 0;
449 btrfs_run_delayed_refs(trans
, root
, updates
);
451 trans
->block_rsv
= rsv
;
453 return should_end_transaction(trans
, root
);
456 static int __btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
457 struct btrfs_root
*root
, int throttle
, int lock
)
459 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
460 struct btrfs_fs_info
*info
= root
->fs_info
;
463 if (--trans
->use_count
) {
464 trans
->block_rsv
= trans
->orig_rsv
;
468 btrfs_trans_release_metadata(trans
, root
);
469 trans
->block_rsv
= NULL
;
471 unsigned long cur
= trans
->delayed_ref_updates
;
472 trans
->delayed_ref_updates
= 0;
474 trans
->transaction
->delayed_refs
.num_heads_ready
> 64) {
475 trans
->delayed_ref_updates
= 0;
476 btrfs_run_delayed_refs(trans
, root
, cur
);
483 if (lock
&& !atomic_read(&root
->fs_info
->open_ioctl_trans
) &&
484 should_end_transaction(trans
, root
)) {
485 trans
->transaction
->blocked
= 1;
489 if (lock
&& cur_trans
->blocked
&& !cur_trans
->in_commit
) {
492 * We may race with somebody else here so end up having
493 * to call end_transaction on ourselves again, so inc
497 return btrfs_commit_transaction(trans
, root
);
499 wake_up_process(info
->transaction_kthread
);
503 WARN_ON(cur_trans
!= info
->running_transaction
);
504 WARN_ON(atomic_read(&cur_trans
->num_writers
) < 1);
505 atomic_dec(&cur_trans
->num_writers
);
508 if (waitqueue_active(&cur_trans
->writer_wait
))
509 wake_up(&cur_trans
->writer_wait
);
510 put_transaction(cur_trans
);
512 if (current
->journal_info
== trans
)
513 current
->journal_info
= NULL
;
514 memset(trans
, 0, sizeof(*trans
));
515 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
518 btrfs_run_delayed_iputs(root
);
523 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
524 struct btrfs_root
*root
)
528 ret
= __btrfs_end_transaction(trans
, root
, 0, 1);
534 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
,
535 struct btrfs_root
*root
)
539 ret
= __btrfs_end_transaction(trans
, root
, 1, 1);
545 int btrfs_end_transaction_nolock(struct btrfs_trans_handle
*trans
,
546 struct btrfs_root
*root
)
550 ret
= __btrfs_end_transaction(trans
, root
, 0, 0);
556 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle
*trans
,
557 struct btrfs_root
*root
)
559 return __btrfs_end_transaction(trans
, root
, 1, 1);
563 * when btree blocks are allocated, they have some corresponding bits set for
564 * them in one of two extent_io trees. This is used to make sure all of
565 * those extents are sent to disk but does not wait on them
567 int btrfs_write_marked_extents(struct btrfs_root
*root
,
568 struct extent_io_tree
*dirty_pages
, int mark
)
572 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
576 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
578 convert_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
, mark
,
580 err
= filemap_fdatawrite_range(mapping
, start
, end
);
592 * when btree blocks are allocated, they have some corresponding bits set for
593 * them in one of two extent_io trees. This is used to make sure all of
594 * those extents are on disk for transaction or log commit. We wait
595 * on all the pages and clear them from the dirty pages state tree
597 int btrfs_wait_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 clear_extent_bits(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
, GFP_NOFS
);
609 err
= filemap_fdatawait_range(mapping
, start
, end
);
621 * when btree blocks are allocated, they have some corresponding bits set for
622 * them in one of two extent_io trees. This is used to make sure all of
623 * those extents are on disk for transaction or log commit
625 int btrfs_write_and_wait_marked_extents(struct btrfs_root
*root
,
626 struct extent_io_tree
*dirty_pages
, int mark
)
631 ret
= btrfs_write_marked_extents(root
, dirty_pages
, mark
);
632 ret2
= btrfs_wait_marked_extents(root
, dirty_pages
, mark
);
641 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
,
642 struct btrfs_root
*root
)
644 if (!trans
|| !trans
->transaction
) {
645 struct inode
*btree_inode
;
646 btree_inode
= root
->fs_info
->btree_inode
;
647 return filemap_write_and_wait(btree_inode
->i_mapping
);
649 return btrfs_write_and_wait_marked_extents(root
,
650 &trans
->transaction
->dirty_pages
,
655 * this is used to update the root pointer in the tree of tree roots.
657 * But, in the case of the extent allocation tree, updating the root
658 * pointer may allocate blocks which may change the root of the extent
661 * So, this loops and repeats and makes sure the cowonly root didn't
662 * change while the root pointer was being updated in the metadata.
664 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
665 struct btrfs_root
*root
)
670 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
672 old_root_used
= btrfs_root_used(&root
->root_item
);
673 btrfs_write_dirty_block_groups(trans
, root
);
676 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
677 if (old_root_bytenr
== root
->node
->start
&&
678 old_root_used
== btrfs_root_used(&root
->root_item
))
681 btrfs_set_root_node(&root
->root_item
, root
->node
);
682 ret
= btrfs_update_root(trans
, tree_root
,
687 old_root_used
= btrfs_root_used(&root
->root_item
);
688 ret
= btrfs_write_dirty_block_groups(trans
, root
);
692 if (root
!= root
->fs_info
->extent_root
)
693 switch_commit_root(root
);
699 * update all the cowonly tree roots on disk
701 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
,
702 struct btrfs_root
*root
)
704 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
705 struct list_head
*next
;
706 struct extent_buffer
*eb
;
709 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
712 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
713 btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
, 0, &eb
);
714 btrfs_tree_unlock(eb
);
715 free_extent_buffer(eb
);
717 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
720 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
721 next
= fs_info
->dirty_cowonly_roots
.next
;
723 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
725 update_cowonly_root(trans
, root
);
728 down_write(&fs_info
->extent_commit_sem
);
729 switch_commit_root(fs_info
->extent_root
);
730 up_write(&fs_info
->extent_commit_sem
);
736 * dead roots are old snapshots that need to be deleted. This allocates
737 * a dirty root struct and adds it into the list of dead roots that need to
740 int btrfs_add_dead_root(struct btrfs_root
*root
)
742 spin_lock(&root
->fs_info
->trans_lock
);
743 list_add(&root
->root_list
, &root
->fs_info
->dead_roots
);
744 spin_unlock(&root
->fs_info
->trans_lock
);
749 * update all the cowonly tree roots on disk
751 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
,
752 struct btrfs_root
*root
)
754 struct btrfs_root
*gang
[8];
755 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
760 spin_lock(&fs_info
->fs_roots_radix_lock
);
762 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
765 BTRFS_ROOT_TRANS_TAG
);
768 for (i
= 0; i
< ret
; i
++) {
770 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
771 (unsigned long)root
->root_key
.objectid
,
772 BTRFS_ROOT_TRANS_TAG
);
773 spin_unlock(&fs_info
->fs_roots_radix_lock
);
775 btrfs_free_log(trans
, root
);
776 btrfs_update_reloc_root(trans
, root
);
777 btrfs_orphan_commit_root(trans
, root
);
779 btrfs_save_ino_cache(root
, trans
);
781 /* see comments in should_cow_block() */
785 if (root
->commit_root
!= root
->node
) {
786 mutex_lock(&root
->fs_commit_mutex
);
787 switch_commit_root(root
);
788 btrfs_unpin_free_ino(root
);
789 mutex_unlock(&root
->fs_commit_mutex
);
791 btrfs_set_root_node(&root
->root_item
,
795 err
= btrfs_update_root(trans
, fs_info
->tree_root
,
798 spin_lock(&fs_info
->fs_roots_radix_lock
);
803 spin_unlock(&fs_info
->fs_roots_radix_lock
);
808 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
809 * otherwise every leaf in the btree is read and defragged.
811 int btrfs_defrag_root(struct btrfs_root
*root
, int cacheonly
)
813 struct btrfs_fs_info
*info
= root
->fs_info
;
814 struct btrfs_trans_handle
*trans
;
818 if (xchg(&root
->defrag_running
, 1))
822 trans
= btrfs_start_transaction(root
, 0);
824 return PTR_ERR(trans
);
826 ret
= btrfs_defrag_leaves(trans
, root
, cacheonly
);
828 nr
= trans
->blocks_used
;
829 btrfs_end_transaction(trans
, root
);
830 btrfs_btree_balance_dirty(info
->tree_root
, nr
);
833 if (btrfs_fs_closing(root
->fs_info
) || ret
!= -EAGAIN
)
836 root
->defrag_running
= 0;
841 * new snapshots need to be created at a very specific time in the
842 * transaction commit. This does the actual creation
844 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
845 struct btrfs_fs_info
*fs_info
,
846 struct btrfs_pending_snapshot
*pending
)
848 struct btrfs_key key
;
849 struct btrfs_root_item
*new_root_item
;
850 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
851 struct btrfs_root
*root
= pending
->root
;
852 struct btrfs_root
*parent_root
;
853 struct btrfs_block_rsv
*rsv
;
854 struct inode
*parent_inode
;
855 struct dentry
*parent
;
856 struct dentry
*dentry
;
857 struct extent_buffer
*tmp
;
858 struct extent_buffer
*old
;
865 rsv
= trans
->block_rsv
;
867 new_root_item
= kmalloc(sizeof(*new_root_item
), GFP_NOFS
);
868 if (!new_root_item
) {
869 pending
->error
= -ENOMEM
;
873 ret
= btrfs_find_free_objectid(tree_root
, &objectid
);
875 pending
->error
= ret
;
879 btrfs_reloc_pre_snapshot(trans
, pending
, &to_reserve
);
881 if (to_reserve
> 0) {
882 ret
= btrfs_block_rsv_add_noflush(root
, &pending
->block_rsv
,
885 pending
->error
= ret
;
890 key
.objectid
= objectid
;
891 key
.offset
= (u64
)-1;
892 key
.type
= BTRFS_ROOT_ITEM_KEY
;
894 trans
->block_rsv
= &pending
->block_rsv
;
896 dentry
= pending
->dentry
;
897 parent
= dget_parent(dentry
);
898 parent_inode
= parent
->d_inode
;
899 parent_root
= BTRFS_I(parent_inode
)->root
;
900 record_root_in_trans(trans
, parent_root
);
903 * insert the directory item
905 ret
= btrfs_set_inode_index(parent_inode
, &index
);
907 ret
= btrfs_insert_dir_item(trans
, parent_root
,
908 dentry
->d_name
.name
, dentry
->d_name
.len
,
910 BTRFS_FT_DIR
, index
);
913 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
914 dentry
->d_name
.len
* 2);
915 ret
= btrfs_update_inode(trans
, parent_root
, parent_inode
);
919 * pull in the delayed directory update
920 * and the delayed inode item
921 * otherwise we corrupt the FS during
924 ret
= btrfs_run_delayed_items(trans
, root
);
927 record_root_in_trans(trans
, root
);
928 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
929 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
930 btrfs_check_and_init_root_item(new_root_item
);
932 root_flags
= btrfs_root_flags(new_root_item
);
933 if (pending
->readonly
)
934 root_flags
|= BTRFS_ROOT_SUBVOL_RDONLY
;
936 root_flags
&= ~BTRFS_ROOT_SUBVOL_RDONLY
;
937 btrfs_set_root_flags(new_root_item
, root_flags
);
939 old
= btrfs_lock_root_node(root
);
940 btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
);
941 btrfs_set_lock_blocking(old
);
943 btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
944 btrfs_tree_unlock(old
);
945 free_extent_buffer(old
);
947 /* see comments in should_cow_block() */
951 btrfs_set_root_node(new_root_item
, tmp
);
952 /* record when the snapshot was created in key.offset */
953 key
.offset
= trans
->transid
;
954 ret
= btrfs_insert_root(trans
, tree_root
, &key
, new_root_item
);
955 btrfs_tree_unlock(tmp
);
956 free_extent_buffer(tmp
);
960 * insert root back/forward references
962 ret
= btrfs_add_root_ref(trans
, tree_root
, objectid
,
963 parent_root
->root_key
.objectid
,
964 btrfs_ino(parent_inode
), index
,
965 dentry
->d_name
.name
, dentry
->d_name
.len
);
969 key
.offset
= (u64
)-1;
970 pending
->snap
= btrfs_read_fs_root_no_name(root
->fs_info
, &key
);
971 BUG_ON(IS_ERR(pending
->snap
));
973 btrfs_reloc_post_snapshot(trans
, pending
);
975 kfree(new_root_item
);
976 trans
->block_rsv
= rsv
;
977 btrfs_block_rsv_release(root
, &pending
->block_rsv
, (u64
)-1);
982 * create all the snapshots we've scheduled for creation
984 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
,
985 struct btrfs_fs_info
*fs_info
)
987 struct btrfs_pending_snapshot
*pending
;
988 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
991 list_for_each_entry(pending
, head
, list
) {
992 ret
= create_pending_snapshot(trans
, fs_info
, pending
);
998 static void update_super_roots(struct btrfs_root
*root
)
1000 struct btrfs_root_item
*root_item
;
1001 struct btrfs_super_block
*super
;
1003 super
= root
->fs_info
->super_copy
;
1005 root_item
= &root
->fs_info
->chunk_root
->root_item
;
1006 super
->chunk_root
= root_item
->bytenr
;
1007 super
->chunk_root_generation
= root_item
->generation
;
1008 super
->chunk_root_level
= root_item
->level
;
1010 root_item
= &root
->fs_info
->tree_root
->root_item
;
1011 super
->root
= root_item
->bytenr
;
1012 super
->generation
= root_item
->generation
;
1013 super
->root_level
= root_item
->level
;
1014 if (btrfs_test_opt(root
, SPACE_CACHE
))
1015 super
->cache_generation
= root_item
->generation
;
1018 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
1021 spin_lock(&info
->trans_lock
);
1022 if (info
->running_transaction
)
1023 ret
= info
->running_transaction
->in_commit
;
1024 spin_unlock(&info
->trans_lock
);
1028 int btrfs_transaction_blocked(struct btrfs_fs_info
*info
)
1031 spin_lock(&info
->trans_lock
);
1032 if (info
->running_transaction
)
1033 ret
= info
->running_transaction
->blocked
;
1034 spin_unlock(&info
->trans_lock
);
1039 * wait for the current transaction commit to start and block subsequent
1042 static void wait_current_trans_commit_start(struct btrfs_root
*root
,
1043 struct btrfs_transaction
*trans
)
1045 wait_event(root
->fs_info
->transaction_blocked_wait
, trans
->in_commit
);
1049 * wait for the current transaction to start and then become unblocked.
1052 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root
*root
,
1053 struct btrfs_transaction
*trans
)
1055 wait_event(root
->fs_info
->transaction_wait
,
1056 trans
->commit_done
|| (trans
->in_commit
&& !trans
->blocked
));
1060 * commit transactions asynchronously. once btrfs_commit_transaction_async
1061 * returns, any subsequent transaction will not be allowed to join.
1063 struct btrfs_async_commit
{
1064 struct btrfs_trans_handle
*newtrans
;
1065 struct btrfs_root
*root
;
1066 struct delayed_work work
;
1069 static void do_async_commit(struct work_struct
*work
)
1071 struct btrfs_async_commit
*ac
=
1072 container_of(work
, struct btrfs_async_commit
, work
.work
);
1074 btrfs_commit_transaction(ac
->newtrans
, ac
->root
);
1078 int btrfs_commit_transaction_async(struct btrfs_trans_handle
*trans
,
1079 struct btrfs_root
*root
,
1080 int wait_for_unblock
)
1082 struct btrfs_async_commit
*ac
;
1083 struct btrfs_transaction
*cur_trans
;
1085 ac
= kmalloc(sizeof(*ac
), GFP_NOFS
);
1089 INIT_DELAYED_WORK(&ac
->work
, do_async_commit
);
1091 ac
->newtrans
= btrfs_join_transaction(root
);
1092 if (IS_ERR(ac
->newtrans
)) {
1093 int err
= PTR_ERR(ac
->newtrans
);
1098 /* take transaction reference */
1099 cur_trans
= trans
->transaction
;
1100 atomic_inc(&cur_trans
->use_count
);
1102 btrfs_end_transaction(trans
, root
);
1103 schedule_delayed_work(&ac
->work
, 0);
1105 /* wait for transaction to start and unblock */
1106 if (wait_for_unblock
)
1107 wait_current_trans_commit_start_and_unblock(root
, cur_trans
);
1109 wait_current_trans_commit_start(root
, cur_trans
);
1111 if (current
->journal_info
== trans
)
1112 current
->journal_info
= NULL
;
1114 put_transaction(cur_trans
);
1119 * btrfs_transaction state sequence:
1120 * in_commit = 0, blocked = 0 (initial)
1121 * in_commit = 1, blocked = 1
1125 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
,
1126 struct btrfs_root
*root
)
1128 unsigned long joined
= 0;
1129 struct btrfs_transaction
*cur_trans
;
1130 struct btrfs_transaction
*prev_trans
= NULL
;
1133 int should_grow
= 0;
1134 unsigned long now
= get_seconds();
1135 int flush_on_commit
= btrfs_test_opt(root
, FLUSHONCOMMIT
);
1137 btrfs_run_ordered_operations(root
, 0);
1139 btrfs_trans_release_metadata(trans
, root
);
1140 trans
->block_rsv
= NULL
;
1142 /* make a pass through all the delayed refs we have so far
1143 * any runnings procs may add more while we are here
1145 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1148 cur_trans
= trans
->transaction
;
1150 * set the flushing flag so procs in this transaction have to
1151 * start sending their work down.
1153 cur_trans
->delayed_refs
.flushing
= 1;
1155 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1158 spin_lock(&cur_trans
->commit_lock
);
1159 if (cur_trans
->in_commit
) {
1160 spin_unlock(&cur_trans
->commit_lock
);
1161 atomic_inc(&cur_trans
->use_count
);
1162 btrfs_end_transaction(trans
, root
);
1164 wait_for_commit(root
, cur_trans
);
1166 put_transaction(cur_trans
);
1171 trans
->transaction
->in_commit
= 1;
1172 trans
->transaction
->blocked
= 1;
1173 spin_unlock(&cur_trans
->commit_lock
);
1174 wake_up(&root
->fs_info
->transaction_blocked_wait
);
1176 spin_lock(&root
->fs_info
->trans_lock
);
1177 if (cur_trans
->list
.prev
!= &root
->fs_info
->trans_list
) {
1178 prev_trans
= list_entry(cur_trans
->list
.prev
,
1179 struct btrfs_transaction
, list
);
1180 if (!prev_trans
->commit_done
) {
1181 atomic_inc(&prev_trans
->use_count
);
1182 spin_unlock(&root
->fs_info
->trans_lock
);
1184 wait_for_commit(root
, prev_trans
);
1186 put_transaction(prev_trans
);
1188 spin_unlock(&root
->fs_info
->trans_lock
);
1191 spin_unlock(&root
->fs_info
->trans_lock
);
1194 if (now
< cur_trans
->start_time
|| now
- cur_trans
->start_time
< 1)
1198 int snap_pending
= 0;
1200 joined
= cur_trans
->num_joined
;
1201 if (!list_empty(&trans
->transaction
->pending_snapshots
))
1204 WARN_ON(cur_trans
!= trans
->transaction
);
1206 if (flush_on_commit
|| snap_pending
) {
1207 btrfs_start_delalloc_inodes(root
, 1);
1208 ret
= btrfs_wait_ordered_extents(root
, 0, 1);
1212 ret
= btrfs_run_delayed_items(trans
, root
);
1216 * rename don't use btrfs_join_transaction, so, once we
1217 * set the transaction to blocked above, we aren't going
1218 * to get any new ordered operations. We can safely run
1219 * it here and no for sure that nothing new will be added
1222 btrfs_run_ordered_operations(root
, 1);
1224 prepare_to_wait(&cur_trans
->writer_wait
, &wait
,
1225 TASK_UNINTERRUPTIBLE
);
1227 if (atomic_read(&cur_trans
->num_writers
) > 1)
1228 schedule_timeout(MAX_SCHEDULE_TIMEOUT
);
1229 else if (should_grow
)
1230 schedule_timeout(1);
1232 finish_wait(&cur_trans
->writer_wait
, &wait
);
1233 } while (atomic_read(&cur_trans
->num_writers
) > 1 ||
1234 (should_grow
&& cur_trans
->num_joined
!= joined
));
1237 * Ok now we need to make sure to block out any other joins while we
1238 * commit the transaction. We could have started a join before setting
1239 * no_join so make sure to wait for num_writers to == 1 again.
1241 spin_lock(&root
->fs_info
->trans_lock
);
1242 root
->fs_info
->trans_no_join
= 1;
1243 spin_unlock(&root
->fs_info
->trans_lock
);
1244 wait_event(cur_trans
->writer_wait
,
1245 atomic_read(&cur_trans
->num_writers
) == 1);
1248 * the reloc mutex makes sure that we stop
1249 * the balancing code from coming in and moving
1250 * extents around in the middle of the commit
1252 mutex_lock(&root
->fs_info
->reloc_mutex
);
1254 ret
= btrfs_run_delayed_items(trans
, root
);
1257 ret
= create_pending_snapshots(trans
, root
->fs_info
);
1260 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1264 * make sure none of the code above managed to slip in a
1267 btrfs_assert_delayed_root_empty(root
);
1269 WARN_ON(cur_trans
!= trans
->transaction
);
1271 btrfs_scrub_pause(root
);
1272 /* btrfs_commit_tree_roots is responsible for getting the
1273 * various roots consistent with each other. Every pointer
1274 * in the tree of tree roots has to point to the most up to date
1275 * root for every subvolume and other tree. So, we have to keep
1276 * the tree logging code from jumping in and changing any
1279 * At this point in the commit, there can't be any tree-log
1280 * writers, but a little lower down we drop the trans mutex
1281 * and let new people in. By holding the tree_log_mutex
1282 * from now until after the super is written, we avoid races
1283 * with the tree-log code.
1285 mutex_lock(&root
->fs_info
->tree_log_mutex
);
1287 ret
= commit_fs_roots(trans
, root
);
1290 /* commit_fs_roots gets rid of all the tree log roots, it is now
1291 * safe to free the root of tree log roots
1293 btrfs_free_log_root_tree(trans
, root
->fs_info
);
1295 ret
= commit_cowonly_roots(trans
, root
);
1298 btrfs_prepare_extent_commit(trans
, root
);
1300 cur_trans
= root
->fs_info
->running_transaction
;
1302 btrfs_set_root_node(&root
->fs_info
->tree_root
->root_item
,
1303 root
->fs_info
->tree_root
->node
);
1304 switch_commit_root(root
->fs_info
->tree_root
);
1306 btrfs_set_root_node(&root
->fs_info
->chunk_root
->root_item
,
1307 root
->fs_info
->chunk_root
->node
);
1308 switch_commit_root(root
->fs_info
->chunk_root
);
1310 update_super_roots(root
);
1312 if (!root
->fs_info
->log_root_recovering
) {
1313 btrfs_set_super_log_root(root
->fs_info
->super_copy
, 0);
1314 btrfs_set_super_log_root_level(root
->fs_info
->super_copy
, 0);
1317 memcpy(root
->fs_info
->super_for_commit
, root
->fs_info
->super_copy
,
1318 sizeof(*root
->fs_info
->super_copy
));
1320 trans
->transaction
->blocked
= 0;
1321 spin_lock(&root
->fs_info
->trans_lock
);
1322 root
->fs_info
->running_transaction
= NULL
;
1323 root
->fs_info
->trans_no_join
= 0;
1324 spin_unlock(&root
->fs_info
->trans_lock
);
1325 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1327 wake_up(&root
->fs_info
->transaction_wait
);
1329 ret
= btrfs_write_and_wait_transaction(trans
, root
);
1331 write_ctree_super(trans
, root
, 0);
1334 * the super is written, we can safely allow the tree-loggers
1335 * to go about their business
1337 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1339 btrfs_finish_extent_commit(trans
, root
);
1341 cur_trans
->commit_done
= 1;
1343 root
->fs_info
->last_trans_committed
= cur_trans
->transid
;
1345 wake_up(&cur_trans
->commit_wait
);
1347 spin_lock(&root
->fs_info
->trans_lock
);
1348 list_del_init(&cur_trans
->list
);
1349 spin_unlock(&root
->fs_info
->trans_lock
);
1351 put_transaction(cur_trans
);
1352 put_transaction(cur_trans
);
1354 trace_btrfs_transaction_commit(root
);
1356 btrfs_scrub_continue(root
);
1358 if (current
->journal_info
== trans
)
1359 current
->journal_info
= NULL
;
1361 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1363 if (current
!= root
->fs_info
->transaction_kthread
)
1364 btrfs_run_delayed_iputs(root
);
1370 * interface function to delete all the snapshots we have scheduled for deletion
1372 int btrfs_clean_old_snapshots(struct btrfs_root
*root
)
1375 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1377 spin_lock(&fs_info
->trans_lock
);
1378 list_splice_init(&fs_info
->dead_roots
, &list
);
1379 spin_unlock(&fs_info
->trans_lock
);
1381 while (!list_empty(&list
)) {
1382 root
= list_entry(list
.next
, struct btrfs_root
, root_list
);
1383 list_del(&root
->root_list
);
1385 btrfs_kill_all_delayed_nodes(root
);
1387 if (btrfs_header_backref_rev(root
->node
) <
1388 BTRFS_MIXED_BACKREF_REV
)
1389 btrfs_drop_snapshot(root
, NULL
, 0);
1391 btrfs_drop_snapshot(root
, NULL
, 1);