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
);
58 if (root
->fs_info
->trans_no_join
) {
60 spin_unlock(&root
->fs_info
->trans_lock
);
65 cur_trans
= root
->fs_info
->running_transaction
;
67 atomic_inc(&cur_trans
->use_count
);
68 atomic_inc(&cur_trans
->num_writers
);
69 cur_trans
->num_joined
++;
70 spin_unlock(&root
->fs_info
->trans_lock
);
73 spin_unlock(&root
->fs_info
->trans_lock
);
75 cur_trans
= kmem_cache_alloc(btrfs_transaction_cachep
, GFP_NOFS
);
78 spin_lock(&root
->fs_info
->trans_lock
);
79 if (root
->fs_info
->running_transaction
) {
80 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
81 cur_trans
= root
->fs_info
->running_transaction
;
82 atomic_inc(&cur_trans
->use_count
);
83 atomic_inc(&cur_trans
->num_writers
);
84 cur_trans
->num_joined
++;
85 spin_unlock(&root
->fs_info
->trans_lock
);
88 atomic_set(&cur_trans
->num_writers
, 1);
89 cur_trans
->num_joined
= 0;
90 init_waitqueue_head(&cur_trans
->writer_wait
);
91 init_waitqueue_head(&cur_trans
->commit_wait
);
92 cur_trans
->in_commit
= 0;
93 cur_trans
->blocked
= 0;
95 * One for this trans handle, one so it will live on until we
96 * commit the transaction.
98 atomic_set(&cur_trans
->use_count
, 2);
99 cur_trans
->commit_done
= 0;
100 cur_trans
->start_time
= get_seconds();
102 cur_trans
->delayed_refs
.root
= RB_ROOT
;
103 cur_trans
->delayed_refs
.num_entries
= 0;
104 cur_trans
->delayed_refs
.num_heads_ready
= 0;
105 cur_trans
->delayed_refs
.num_heads
= 0;
106 cur_trans
->delayed_refs
.flushing
= 0;
107 cur_trans
->delayed_refs
.run_delayed_start
= 0;
108 spin_lock_init(&cur_trans
->commit_lock
);
109 spin_lock_init(&cur_trans
->delayed_refs
.lock
);
111 INIT_LIST_HEAD(&cur_trans
->pending_snapshots
);
112 list_add_tail(&cur_trans
->list
, &root
->fs_info
->trans_list
);
113 extent_io_tree_init(&cur_trans
->dirty_pages
,
114 root
->fs_info
->btree_inode
->i_mapping
);
115 root
->fs_info
->generation
++;
116 cur_trans
->transid
= root
->fs_info
->generation
;
117 root
->fs_info
->running_transaction
= cur_trans
;
118 spin_unlock(&root
->fs_info
->trans_lock
);
124 * this does all the record keeping required to make sure that a reference
125 * counted root is properly recorded in a given transaction. This is required
126 * to make sure the old root from before we joined the transaction is deleted
127 * when the transaction commits
129 static int record_root_in_trans(struct btrfs_trans_handle
*trans
,
130 struct btrfs_root
*root
)
132 if (root
->ref_cows
&& root
->last_trans
< trans
->transid
) {
133 WARN_ON(root
== root
->fs_info
->extent_root
);
134 WARN_ON(root
->commit_root
!= root
->node
);
137 * see below for in_trans_setup usage rules
138 * we have the reloc mutex held now, so there
139 * is only one writer in this function
141 root
->in_trans_setup
= 1;
143 /* make sure readers find in_trans_setup before
144 * they find our root->last_trans update
148 spin_lock(&root
->fs_info
->fs_roots_radix_lock
);
149 if (root
->last_trans
== trans
->transid
) {
150 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
153 radix_tree_tag_set(&root
->fs_info
->fs_roots_radix
,
154 (unsigned long)root
->root_key
.objectid
,
155 BTRFS_ROOT_TRANS_TAG
);
156 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
157 root
->last_trans
= trans
->transid
;
159 /* this is pretty tricky. We don't want to
160 * take the relocation lock in btrfs_record_root_in_trans
161 * unless we're really doing the first setup for this root in
164 * Normally we'd use root->last_trans as a flag to decide
165 * if we want to take the expensive mutex.
167 * But, we have to set root->last_trans before we
168 * init the relocation root, otherwise, we trip over warnings
169 * in ctree.c. The solution used here is to flag ourselves
170 * with root->in_trans_setup. When this is 1, we're still
171 * fixing up the reloc trees and everyone must wait.
173 * When this is zero, they can trust root->last_trans and fly
174 * through btrfs_record_root_in_trans without having to take the
175 * lock. smp_wmb() makes sure that all the writes above are
176 * done before we pop in the zero below
178 btrfs_init_reloc_root(trans
, root
);
180 root
->in_trans_setup
= 0;
186 int btrfs_record_root_in_trans(struct btrfs_trans_handle
*trans
,
187 struct btrfs_root
*root
)
193 * see record_root_in_trans for comments about in_trans_setup usage
197 if (root
->last_trans
== trans
->transid
&&
198 !root
->in_trans_setup
)
201 mutex_lock(&root
->fs_info
->reloc_mutex
);
202 record_root_in_trans(trans
, root
);
203 mutex_unlock(&root
->fs_info
->reloc_mutex
);
208 /* wait for commit against the current transaction to become unblocked
209 * when this is done, it is safe to start a new transaction, but the current
210 * transaction might not be fully on disk.
212 static void wait_current_trans(struct btrfs_root
*root
)
214 struct btrfs_transaction
*cur_trans
;
216 spin_lock(&root
->fs_info
->trans_lock
);
217 cur_trans
= root
->fs_info
->running_transaction
;
218 if (cur_trans
&& cur_trans
->blocked
) {
220 atomic_inc(&cur_trans
->use_count
);
221 spin_unlock(&root
->fs_info
->trans_lock
);
223 prepare_to_wait(&root
->fs_info
->transaction_wait
, &wait
,
224 TASK_UNINTERRUPTIBLE
);
225 if (!cur_trans
->blocked
)
229 finish_wait(&root
->fs_info
->transaction_wait
, &wait
);
230 put_transaction(cur_trans
);
232 spin_unlock(&root
->fs_info
->trans_lock
);
236 enum btrfs_trans_type
{
243 static int may_wait_transaction(struct btrfs_root
*root
, int type
)
245 if (root
->fs_info
->log_root_recovering
)
248 if (type
== TRANS_USERSPACE
)
251 if (type
== TRANS_START
&&
252 !atomic_read(&root
->fs_info
->open_ioctl_trans
))
258 static struct btrfs_trans_handle
*start_transaction(struct btrfs_root
*root
,
259 u64 num_items
, int type
)
261 struct btrfs_trans_handle
*h
;
262 struct btrfs_transaction
*cur_trans
;
266 if (root
->fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
)
267 return ERR_PTR(-EROFS
);
269 if (current
->journal_info
) {
270 WARN_ON(type
!= TRANS_JOIN
&& type
!= TRANS_JOIN_NOLOCK
);
271 h
= current
->journal_info
;
273 h
->orig_rsv
= h
->block_rsv
;
279 * Do the reservation before we join the transaction so we can do all
280 * the appropriate flushing if need be.
282 if (num_items
> 0 && root
!= root
->fs_info
->chunk_root
) {
283 num_bytes
= btrfs_calc_trans_metadata_size(root
, num_items
);
284 ret
= btrfs_block_rsv_add(NULL
, root
,
285 &root
->fs_info
->trans_block_rsv
,
291 h
= kmem_cache_alloc(btrfs_trans_handle_cachep
, GFP_NOFS
);
293 return ERR_PTR(-ENOMEM
);
295 if (may_wait_transaction(root
, type
))
296 wait_current_trans(root
);
299 ret
= join_transaction(root
, type
== TRANS_JOIN_NOLOCK
);
301 wait_current_trans(root
);
302 } while (ret
== -EBUSY
);
305 kmem_cache_free(btrfs_trans_handle_cachep
, h
);
309 cur_trans
= root
->fs_info
->running_transaction
;
311 h
->transid
= cur_trans
->transid
;
312 h
->transaction
= cur_trans
;
314 h
->bytes_reserved
= 0;
315 h
->delayed_ref_updates
= 0;
321 if (cur_trans
->blocked
&& may_wait_transaction(root
, type
)) {
322 btrfs_commit_transaction(h
, root
);
327 h
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
328 h
->bytes_reserved
= num_bytes
;
332 btrfs_record_root_in_trans(h
, root
);
334 if (!current
->journal_info
&& type
!= TRANS_USERSPACE
)
335 current
->journal_info
= h
;
339 struct btrfs_trans_handle
*btrfs_start_transaction(struct btrfs_root
*root
,
342 return start_transaction(root
, num_items
, TRANS_START
);
344 struct btrfs_trans_handle
*btrfs_join_transaction(struct btrfs_root
*root
)
346 return start_transaction(root
, 0, TRANS_JOIN
);
349 struct btrfs_trans_handle
*btrfs_join_transaction_nolock(struct btrfs_root
*root
)
351 return start_transaction(root
, 0, TRANS_JOIN_NOLOCK
);
354 struct btrfs_trans_handle
*btrfs_start_ioctl_transaction(struct btrfs_root
*root
)
356 return start_transaction(root
, 0, TRANS_USERSPACE
);
359 /* wait for a transaction commit to be fully complete */
360 static noinline
int wait_for_commit(struct btrfs_root
*root
,
361 struct btrfs_transaction
*commit
)
364 while (!commit
->commit_done
) {
365 prepare_to_wait(&commit
->commit_wait
, &wait
,
366 TASK_UNINTERRUPTIBLE
);
367 if (commit
->commit_done
)
371 finish_wait(&commit
->commit_wait
, &wait
);
375 int btrfs_wait_for_commit(struct btrfs_root
*root
, u64 transid
)
377 struct btrfs_transaction
*cur_trans
= NULL
, *t
;
382 if (transid
<= root
->fs_info
->last_trans_committed
)
385 /* find specified transaction */
386 spin_lock(&root
->fs_info
->trans_lock
);
387 list_for_each_entry(t
, &root
->fs_info
->trans_list
, list
) {
388 if (t
->transid
== transid
) {
390 atomic_inc(&cur_trans
->use_count
);
393 if (t
->transid
> transid
)
396 spin_unlock(&root
->fs_info
->trans_lock
);
399 goto out
; /* bad transid */
401 /* find newest transaction that is committing | committed */
402 spin_lock(&root
->fs_info
->trans_lock
);
403 list_for_each_entry_reverse(t
, &root
->fs_info
->trans_list
,
409 atomic_inc(&cur_trans
->use_count
);
413 spin_unlock(&root
->fs_info
->trans_lock
);
415 goto out
; /* nothing committing|committed */
418 wait_for_commit(root
, cur_trans
);
420 put_transaction(cur_trans
);
426 void btrfs_throttle(struct btrfs_root
*root
)
428 if (!atomic_read(&root
->fs_info
->open_ioctl_trans
))
429 wait_current_trans(root
);
432 static int should_end_transaction(struct btrfs_trans_handle
*trans
,
433 struct btrfs_root
*root
)
436 ret
= btrfs_block_rsv_check(trans
, root
,
437 &root
->fs_info
->global_block_rsv
, 0, 5);
441 int btrfs_should_end_transaction(struct btrfs_trans_handle
*trans
,
442 struct btrfs_root
*root
)
444 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
448 if (cur_trans
->blocked
|| cur_trans
->delayed_refs
.flushing
)
451 updates
= trans
->delayed_ref_updates
;
452 trans
->delayed_ref_updates
= 0;
454 btrfs_run_delayed_refs(trans
, root
, updates
);
456 return should_end_transaction(trans
, root
);
459 static int __btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
460 struct btrfs_root
*root
, int throttle
, int lock
)
462 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
463 struct btrfs_fs_info
*info
= root
->fs_info
;
466 if (--trans
->use_count
) {
467 trans
->block_rsv
= trans
->orig_rsv
;
472 unsigned long cur
= trans
->delayed_ref_updates
;
473 trans
->delayed_ref_updates
= 0;
475 trans
->transaction
->delayed_refs
.num_heads_ready
> 64) {
476 trans
->delayed_ref_updates
= 0;
479 * do a full flush if the transaction is trying
482 if (trans
->transaction
->delayed_refs
.flushing
)
484 btrfs_run_delayed_refs(trans
, root
, cur
);
491 btrfs_trans_release_metadata(trans
, root
);
493 if (lock
&& !atomic_read(&root
->fs_info
->open_ioctl_trans
) &&
494 should_end_transaction(trans
, root
)) {
495 trans
->transaction
->blocked
= 1;
499 if (lock
&& cur_trans
->blocked
&& !cur_trans
->in_commit
) {
502 * We may race with somebody else here so end up having
503 * to call end_transaction on ourselves again, so inc
507 return btrfs_commit_transaction(trans
, root
);
509 wake_up_process(info
->transaction_kthread
);
513 WARN_ON(cur_trans
!= info
->running_transaction
);
514 WARN_ON(atomic_read(&cur_trans
->num_writers
) < 1);
515 atomic_dec(&cur_trans
->num_writers
);
518 if (waitqueue_active(&cur_trans
->writer_wait
))
519 wake_up(&cur_trans
->writer_wait
);
520 put_transaction(cur_trans
);
522 if (current
->journal_info
== trans
)
523 current
->journal_info
= NULL
;
524 memset(trans
, 0, sizeof(*trans
));
525 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
528 btrfs_run_delayed_iputs(root
);
533 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
534 struct btrfs_root
*root
)
538 ret
= __btrfs_end_transaction(trans
, root
, 0, 1);
544 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
,
545 struct btrfs_root
*root
)
549 ret
= __btrfs_end_transaction(trans
, root
, 1, 1);
555 int btrfs_end_transaction_nolock(struct btrfs_trans_handle
*trans
,
556 struct btrfs_root
*root
)
560 ret
= __btrfs_end_transaction(trans
, root
, 0, 0);
566 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle
*trans
,
567 struct btrfs_root
*root
)
569 return __btrfs_end_transaction(trans
, root
, 1, 1);
573 * when btree blocks are allocated, they have some corresponding bits set for
574 * them in one of two extent_io trees. This is used to make sure all of
575 * those extents are sent to disk but does not wait on them
577 int btrfs_write_marked_extents(struct btrfs_root
*root
,
578 struct extent_io_tree
*dirty_pages
, int mark
)
584 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
590 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
594 while (start
<= end
) {
597 index
= start
>> PAGE_CACHE_SHIFT
;
598 start
= (u64
)(index
+ 1) << PAGE_CACHE_SHIFT
;
599 page
= find_get_page(btree_inode
->i_mapping
, index
);
603 btree_lock_page_hook(page
);
604 if (!page
->mapping
) {
606 page_cache_release(page
);
610 if (PageWriteback(page
)) {
612 wait_on_page_writeback(page
);
615 page_cache_release(page
);
619 err
= write_one_page(page
, 0);
622 page_cache_release(page
);
631 * when btree blocks are allocated, they have some corresponding bits set for
632 * them in one of two extent_io trees. This is used to make sure all of
633 * those extents are on disk for transaction or log commit. We wait
634 * on all the pages and clear them from the dirty pages state tree
636 int btrfs_wait_marked_extents(struct btrfs_root
*root
,
637 struct extent_io_tree
*dirty_pages
, int mark
)
643 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
649 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
654 clear_extent_bits(dirty_pages
, start
, end
, mark
, GFP_NOFS
);
655 while (start
<= end
) {
656 index
= start
>> PAGE_CACHE_SHIFT
;
657 start
= (u64
)(index
+ 1) << PAGE_CACHE_SHIFT
;
658 page
= find_get_page(btree_inode
->i_mapping
, index
);
661 if (PageDirty(page
)) {
662 btree_lock_page_hook(page
);
663 wait_on_page_writeback(page
);
664 err
= write_one_page(page
, 0);
668 wait_on_page_writeback(page
);
669 page_cache_release(page
);
679 * when btree blocks are allocated, they have some corresponding bits set for
680 * them in one of two extent_io trees. This is used to make sure all of
681 * those extents are on disk for transaction or log commit
683 int btrfs_write_and_wait_marked_extents(struct btrfs_root
*root
,
684 struct extent_io_tree
*dirty_pages
, int mark
)
689 ret
= btrfs_write_marked_extents(root
, dirty_pages
, mark
);
690 ret2
= btrfs_wait_marked_extents(root
, dirty_pages
, mark
);
694 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
,
695 struct btrfs_root
*root
)
697 if (!trans
|| !trans
->transaction
) {
698 struct inode
*btree_inode
;
699 btree_inode
= root
->fs_info
->btree_inode
;
700 return filemap_write_and_wait(btree_inode
->i_mapping
);
702 return btrfs_write_and_wait_marked_extents(root
,
703 &trans
->transaction
->dirty_pages
,
708 * this is used to update the root pointer in the tree of tree roots.
710 * But, in the case of the extent allocation tree, updating the root
711 * pointer may allocate blocks which may change the root of the extent
714 * So, this loops and repeats and makes sure the cowonly root didn't
715 * change while the root pointer was being updated in the metadata.
717 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
718 struct btrfs_root
*root
)
723 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
725 old_root_used
= btrfs_root_used(&root
->root_item
);
726 btrfs_write_dirty_block_groups(trans
, root
);
729 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
730 if (old_root_bytenr
== root
->node
->start
&&
731 old_root_used
== btrfs_root_used(&root
->root_item
))
734 btrfs_set_root_node(&root
->root_item
, root
->node
);
735 ret
= btrfs_update_root(trans
, tree_root
,
740 old_root_used
= btrfs_root_used(&root
->root_item
);
741 ret
= btrfs_write_dirty_block_groups(trans
, root
);
745 if (root
!= root
->fs_info
->extent_root
)
746 switch_commit_root(root
);
752 * update all the cowonly tree roots on disk
754 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
,
755 struct btrfs_root
*root
)
757 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
758 struct list_head
*next
;
759 struct extent_buffer
*eb
;
762 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
765 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
766 btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
, 0, &eb
);
767 btrfs_tree_unlock(eb
);
768 free_extent_buffer(eb
);
770 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
773 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
774 next
= fs_info
->dirty_cowonly_roots
.next
;
776 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
778 update_cowonly_root(trans
, root
);
781 down_write(&fs_info
->extent_commit_sem
);
782 switch_commit_root(fs_info
->extent_root
);
783 up_write(&fs_info
->extent_commit_sem
);
789 * dead roots are old snapshots that need to be deleted. This allocates
790 * a dirty root struct and adds it into the list of dead roots that need to
793 int btrfs_add_dead_root(struct btrfs_root
*root
)
795 spin_lock(&root
->fs_info
->trans_lock
);
796 list_add(&root
->root_list
, &root
->fs_info
->dead_roots
);
797 spin_unlock(&root
->fs_info
->trans_lock
);
802 * update all the cowonly tree roots on disk
804 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
,
805 struct btrfs_root
*root
)
807 struct btrfs_root
*gang
[8];
808 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
813 spin_lock(&fs_info
->fs_roots_radix_lock
);
815 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
818 BTRFS_ROOT_TRANS_TAG
);
821 for (i
= 0; i
< ret
; i
++) {
823 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
824 (unsigned long)root
->root_key
.objectid
,
825 BTRFS_ROOT_TRANS_TAG
);
826 spin_unlock(&fs_info
->fs_roots_radix_lock
);
828 btrfs_free_log(trans
, root
);
829 btrfs_update_reloc_root(trans
, root
);
830 btrfs_orphan_commit_root(trans
, root
);
832 btrfs_save_ino_cache(root
, trans
);
834 if (root
->commit_root
!= root
->node
) {
835 mutex_lock(&root
->fs_commit_mutex
);
836 switch_commit_root(root
);
837 btrfs_unpin_free_ino(root
);
838 mutex_unlock(&root
->fs_commit_mutex
);
840 btrfs_set_root_node(&root
->root_item
,
844 err
= btrfs_update_root(trans
, fs_info
->tree_root
,
847 spin_lock(&fs_info
->fs_roots_radix_lock
);
852 spin_unlock(&fs_info
->fs_roots_radix_lock
);
857 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
858 * otherwise every leaf in the btree is read and defragged.
860 int btrfs_defrag_root(struct btrfs_root
*root
, int cacheonly
)
862 struct btrfs_fs_info
*info
= root
->fs_info
;
863 struct btrfs_trans_handle
*trans
;
867 if (xchg(&root
->defrag_running
, 1))
871 trans
= btrfs_start_transaction(root
, 0);
873 return PTR_ERR(trans
);
875 ret
= btrfs_defrag_leaves(trans
, root
, cacheonly
);
877 nr
= trans
->blocks_used
;
878 btrfs_end_transaction(trans
, root
);
879 btrfs_btree_balance_dirty(info
->tree_root
, nr
);
882 if (btrfs_fs_closing(root
->fs_info
) || ret
!= -EAGAIN
)
885 root
->defrag_running
= 0;
890 * new snapshots need to be created at a very specific time in the
891 * transaction commit. This does the actual creation
893 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
894 struct btrfs_fs_info
*fs_info
,
895 struct btrfs_pending_snapshot
*pending
)
897 struct btrfs_key key
;
898 struct btrfs_root_item
*new_root_item
;
899 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
900 struct btrfs_root
*root
= pending
->root
;
901 struct btrfs_root
*parent_root
;
902 struct inode
*parent_inode
;
903 struct dentry
*parent
;
904 struct dentry
*dentry
;
905 struct extent_buffer
*tmp
;
906 struct extent_buffer
*old
;
913 new_root_item
= kmalloc(sizeof(*new_root_item
), GFP_NOFS
);
914 if (!new_root_item
) {
915 pending
->error
= -ENOMEM
;
919 ret
= btrfs_find_free_objectid(tree_root
, &objectid
);
921 pending
->error
= ret
;
925 btrfs_reloc_pre_snapshot(trans
, pending
, &to_reserve
);
926 btrfs_orphan_pre_snapshot(trans
, pending
, &to_reserve
);
928 if (to_reserve
> 0) {
929 ret
= btrfs_block_rsv_add(trans
, root
, &pending
->block_rsv
,
932 pending
->error
= ret
;
937 key
.objectid
= objectid
;
938 key
.offset
= (u64
)-1;
939 key
.type
= BTRFS_ROOT_ITEM_KEY
;
941 trans
->block_rsv
= &pending
->block_rsv
;
943 dentry
= pending
->dentry
;
944 parent
= dget_parent(dentry
);
945 parent_inode
= parent
->d_inode
;
946 parent_root
= BTRFS_I(parent_inode
)->root
;
947 record_root_in_trans(trans
, parent_root
);
950 * insert the directory item
952 ret
= btrfs_set_inode_index(parent_inode
, &index
);
954 ret
= btrfs_insert_dir_item(trans
, parent_root
,
955 dentry
->d_name
.name
, dentry
->d_name
.len
,
957 BTRFS_FT_DIR
, index
);
960 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
961 dentry
->d_name
.len
* 2);
962 ret
= btrfs_update_inode(trans
, parent_root
, parent_inode
);
966 * pull in the delayed directory update
967 * and the delayed inode item
968 * otherwise we corrupt the FS during
971 ret
= btrfs_run_delayed_items(trans
, root
);
974 record_root_in_trans(trans
, root
);
975 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
976 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
977 btrfs_check_and_init_root_item(new_root_item
);
979 root_flags
= btrfs_root_flags(new_root_item
);
980 if (pending
->readonly
)
981 root_flags
|= BTRFS_ROOT_SUBVOL_RDONLY
;
983 root_flags
&= ~BTRFS_ROOT_SUBVOL_RDONLY
;
984 btrfs_set_root_flags(new_root_item
, root_flags
);
986 old
= btrfs_lock_root_node(root
);
987 btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
);
988 btrfs_set_lock_blocking(old
);
990 btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
991 btrfs_tree_unlock(old
);
992 free_extent_buffer(old
);
994 btrfs_set_root_node(new_root_item
, tmp
);
995 /* record when the snapshot was created in key.offset */
996 key
.offset
= trans
->transid
;
997 ret
= btrfs_insert_root(trans
, tree_root
, &key
, new_root_item
);
998 btrfs_tree_unlock(tmp
);
999 free_extent_buffer(tmp
);
1003 * insert root back/forward references
1005 ret
= btrfs_add_root_ref(trans
, tree_root
, objectid
,
1006 parent_root
->root_key
.objectid
,
1007 btrfs_ino(parent_inode
), index
,
1008 dentry
->d_name
.name
, dentry
->d_name
.len
);
1012 key
.offset
= (u64
)-1;
1013 pending
->snap
= btrfs_read_fs_root_no_name(root
->fs_info
, &key
);
1014 BUG_ON(IS_ERR(pending
->snap
));
1016 btrfs_reloc_post_snapshot(trans
, pending
);
1017 btrfs_orphan_post_snapshot(trans
, pending
);
1019 kfree(new_root_item
);
1020 btrfs_block_rsv_release(root
, &pending
->block_rsv
, (u64
)-1);
1025 * create all the snapshots we've scheduled for creation
1027 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
,
1028 struct btrfs_fs_info
*fs_info
)
1030 struct btrfs_pending_snapshot
*pending
;
1031 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
1034 list_for_each_entry(pending
, head
, list
) {
1035 ret
= create_pending_snapshot(trans
, fs_info
, pending
);
1041 static void update_super_roots(struct btrfs_root
*root
)
1043 struct btrfs_root_item
*root_item
;
1044 struct btrfs_super_block
*super
;
1046 super
= &root
->fs_info
->super_copy
;
1048 root_item
= &root
->fs_info
->chunk_root
->root_item
;
1049 super
->chunk_root
= root_item
->bytenr
;
1050 super
->chunk_root_generation
= root_item
->generation
;
1051 super
->chunk_root_level
= root_item
->level
;
1053 root_item
= &root
->fs_info
->tree_root
->root_item
;
1054 super
->root
= root_item
->bytenr
;
1055 super
->generation
= root_item
->generation
;
1056 super
->root_level
= root_item
->level
;
1057 if (super
->cache_generation
!= 0 || btrfs_test_opt(root
, SPACE_CACHE
))
1058 super
->cache_generation
= root_item
->generation
;
1061 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
1064 spin_lock(&info
->trans_lock
);
1065 if (info
->running_transaction
)
1066 ret
= info
->running_transaction
->in_commit
;
1067 spin_unlock(&info
->trans_lock
);
1071 int btrfs_transaction_blocked(struct btrfs_fs_info
*info
)
1074 spin_lock(&info
->trans_lock
);
1075 if (info
->running_transaction
)
1076 ret
= info
->running_transaction
->blocked
;
1077 spin_unlock(&info
->trans_lock
);
1082 * wait for the current transaction commit to start and block subsequent
1085 static void wait_current_trans_commit_start(struct btrfs_root
*root
,
1086 struct btrfs_transaction
*trans
)
1090 if (trans
->in_commit
)
1094 prepare_to_wait(&root
->fs_info
->transaction_blocked_wait
, &wait
,
1095 TASK_UNINTERRUPTIBLE
);
1096 if (trans
->in_commit
) {
1097 finish_wait(&root
->fs_info
->transaction_blocked_wait
,
1102 finish_wait(&root
->fs_info
->transaction_blocked_wait
, &wait
);
1107 * wait for the current transaction to start and then become unblocked.
1110 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root
*root
,
1111 struct btrfs_transaction
*trans
)
1115 if (trans
->commit_done
|| (trans
->in_commit
&& !trans
->blocked
))
1119 prepare_to_wait(&root
->fs_info
->transaction_wait
, &wait
,
1120 TASK_UNINTERRUPTIBLE
);
1121 if (trans
->commit_done
||
1122 (trans
->in_commit
&& !trans
->blocked
)) {
1123 finish_wait(&root
->fs_info
->transaction_wait
,
1128 finish_wait(&root
->fs_info
->transaction_wait
,
1134 * commit transactions asynchronously. once btrfs_commit_transaction_async
1135 * returns, any subsequent transaction will not be allowed to join.
1137 struct btrfs_async_commit
{
1138 struct btrfs_trans_handle
*newtrans
;
1139 struct btrfs_root
*root
;
1140 struct delayed_work work
;
1143 static void do_async_commit(struct work_struct
*work
)
1145 struct btrfs_async_commit
*ac
=
1146 container_of(work
, struct btrfs_async_commit
, work
.work
);
1148 btrfs_commit_transaction(ac
->newtrans
, ac
->root
);
1152 int btrfs_commit_transaction_async(struct btrfs_trans_handle
*trans
,
1153 struct btrfs_root
*root
,
1154 int wait_for_unblock
)
1156 struct btrfs_async_commit
*ac
;
1157 struct btrfs_transaction
*cur_trans
;
1159 ac
= kmalloc(sizeof(*ac
), GFP_NOFS
);
1163 INIT_DELAYED_WORK(&ac
->work
, do_async_commit
);
1165 ac
->newtrans
= btrfs_join_transaction(root
);
1166 if (IS_ERR(ac
->newtrans
)) {
1167 int err
= PTR_ERR(ac
->newtrans
);
1172 /* take transaction reference */
1173 cur_trans
= trans
->transaction
;
1174 atomic_inc(&cur_trans
->use_count
);
1176 btrfs_end_transaction(trans
, root
);
1177 schedule_delayed_work(&ac
->work
, 0);
1179 /* wait for transaction to start and unblock */
1180 if (wait_for_unblock
)
1181 wait_current_trans_commit_start_and_unblock(root
, cur_trans
);
1183 wait_current_trans_commit_start(root
, cur_trans
);
1185 if (current
->journal_info
== trans
)
1186 current
->journal_info
= NULL
;
1188 put_transaction(cur_trans
);
1193 * btrfs_transaction state sequence:
1194 * in_commit = 0, blocked = 0 (initial)
1195 * in_commit = 1, blocked = 1
1199 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
,
1200 struct btrfs_root
*root
)
1202 unsigned long joined
= 0;
1203 struct btrfs_transaction
*cur_trans
;
1204 struct btrfs_transaction
*prev_trans
= NULL
;
1207 int should_grow
= 0;
1208 unsigned long now
= get_seconds();
1209 int flush_on_commit
= btrfs_test_opt(root
, FLUSHONCOMMIT
);
1211 btrfs_run_ordered_operations(root
, 0);
1213 /* make a pass through all the delayed refs we have so far
1214 * any runnings procs may add more while we are here
1216 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1219 btrfs_trans_release_metadata(trans
, root
);
1221 cur_trans
= trans
->transaction
;
1223 * set the flushing flag so procs in this transaction have to
1224 * start sending their work down.
1226 cur_trans
->delayed_refs
.flushing
= 1;
1228 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1231 spin_lock(&cur_trans
->commit_lock
);
1232 if (cur_trans
->in_commit
) {
1233 spin_unlock(&cur_trans
->commit_lock
);
1234 atomic_inc(&cur_trans
->use_count
);
1235 btrfs_end_transaction(trans
, root
);
1237 ret
= wait_for_commit(root
, cur_trans
);
1240 put_transaction(cur_trans
);
1245 trans
->transaction
->in_commit
= 1;
1246 trans
->transaction
->blocked
= 1;
1247 spin_unlock(&cur_trans
->commit_lock
);
1248 wake_up(&root
->fs_info
->transaction_blocked_wait
);
1250 spin_lock(&root
->fs_info
->trans_lock
);
1251 if (cur_trans
->list
.prev
!= &root
->fs_info
->trans_list
) {
1252 prev_trans
= list_entry(cur_trans
->list
.prev
,
1253 struct btrfs_transaction
, list
);
1254 if (!prev_trans
->commit_done
) {
1255 atomic_inc(&prev_trans
->use_count
);
1256 spin_unlock(&root
->fs_info
->trans_lock
);
1258 wait_for_commit(root
, prev_trans
);
1260 put_transaction(prev_trans
);
1262 spin_unlock(&root
->fs_info
->trans_lock
);
1265 spin_unlock(&root
->fs_info
->trans_lock
);
1268 if (now
< cur_trans
->start_time
|| now
- cur_trans
->start_time
< 1)
1272 int snap_pending
= 0;
1274 joined
= cur_trans
->num_joined
;
1275 if (!list_empty(&trans
->transaction
->pending_snapshots
))
1278 WARN_ON(cur_trans
!= trans
->transaction
);
1280 if (flush_on_commit
|| snap_pending
) {
1281 btrfs_start_delalloc_inodes(root
, 1);
1282 ret
= btrfs_wait_ordered_extents(root
, 0, 1);
1286 ret
= btrfs_run_delayed_items(trans
, root
);
1290 * rename don't use btrfs_join_transaction, so, once we
1291 * set the transaction to blocked above, we aren't going
1292 * to get any new ordered operations. We can safely run
1293 * it here and no for sure that nothing new will be added
1296 btrfs_run_ordered_operations(root
, 1);
1298 prepare_to_wait(&cur_trans
->writer_wait
, &wait
,
1299 TASK_UNINTERRUPTIBLE
);
1301 if (atomic_read(&cur_trans
->num_writers
) > 1)
1302 schedule_timeout(MAX_SCHEDULE_TIMEOUT
);
1303 else if (should_grow
)
1304 schedule_timeout(1);
1306 finish_wait(&cur_trans
->writer_wait
, &wait
);
1307 } while (atomic_read(&cur_trans
->num_writers
) > 1 ||
1308 (should_grow
&& cur_trans
->num_joined
!= joined
));
1311 * Ok now we need to make sure to block out any other joins while we
1312 * commit the transaction. We could have started a join before setting
1313 * no_join so make sure to wait for num_writers to == 1 again.
1315 spin_lock(&root
->fs_info
->trans_lock
);
1316 root
->fs_info
->trans_no_join
= 1;
1317 spin_unlock(&root
->fs_info
->trans_lock
);
1318 wait_event(cur_trans
->writer_wait
,
1319 atomic_read(&cur_trans
->num_writers
) == 1);
1322 * the reloc mutex makes sure that we stop
1323 * the balancing code from coming in and moving
1324 * extents around in the middle of the commit
1326 mutex_lock(&root
->fs_info
->reloc_mutex
);
1328 ret
= btrfs_run_delayed_items(trans
, root
);
1331 ret
= create_pending_snapshots(trans
, root
->fs_info
);
1334 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1338 * make sure none of the code above managed to slip in a
1341 btrfs_assert_delayed_root_empty(root
);
1343 WARN_ON(cur_trans
!= trans
->transaction
);
1345 btrfs_scrub_pause(root
);
1346 /* btrfs_commit_tree_roots is responsible for getting the
1347 * various roots consistent with each other. Every pointer
1348 * in the tree of tree roots has to point to the most up to date
1349 * root for every subvolume and other tree. So, we have to keep
1350 * the tree logging code from jumping in and changing any
1353 * At this point in the commit, there can't be any tree-log
1354 * writers, but a little lower down we drop the trans mutex
1355 * and let new people in. By holding the tree_log_mutex
1356 * from now until after the super is written, we avoid races
1357 * with the tree-log code.
1359 mutex_lock(&root
->fs_info
->tree_log_mutex
);
1361 ret
= commit_fs_roots(trans
, root
);
1364 /* commit_fs_roots gets rid of all the tree log roots, it is now
1365 * safe to free the root of tree log roots
1367 btrfs_free_log_root_tree(trans
, root
->fs_info
);
1369 ret
= commit_cowonly_roots(trans
, root
);
1372 btrfs_prepare_extent_commit(trans
, root
);
1374 cur_trans
= root
->fs_info
->running_transaction
;
1376 btrfs_set_root_node(&root
->fs_info
->tree_root
->root_item
,
1377 root
->fs_info
->tree_root
->node
);
1378 switch_commit_root(root
->fs_info
->tree_root
);
1380 btrfs_set_root_node(&root
->fs_info
->chunk_root
->root_item
,
1381 root
->fs_info
->chunk_root
->node
);
1382 switch_commit_root(root
->fs_info
->chunk_root
);
1384 update_super_roots(root
);
1386 if (!root
->fs_info
->log_root_recovering
) {
1387 btrfs_set_super_log_root(&root
->fs_info
->super_copy
, 0);
1388 btrfs_set_super_log_root_level(&root
->fs_info
->super_copy
, 0);
1391 memcpy(&root
->fs_info
->super_for_commit
, &root
->fs_info
->super_copy
,
1392 sizeof(root
->fs_info
->super_copy
));
1394 trans
->transaction
->blocked
= 0;
1395 spin_lock(&root
->fs_info
->trans_lock
);
1396 root
->fs_info
->running_transaction
= NULL
;
1397 root
->fs_info
->trans_no_join
= 0;
1398 spin_unlock(&root
->fs_info
->trans_lock
);
1399 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1401 wake_up(&root
->fs_info
->transaction_wait
);
1403 ret
= btrfs_write_and_wait_transaction(trans
, root
);
1405 write_ctree_super(trans
, root
, 0);
1408 * the super is written, we can safely allow the tree-loggers
1409 * to go about their business
1411 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1413 btrfs_finish_extent_commit(trans
, root
);
1415 cur_trans
->commit_done
= 1;
1417 root
->fs_info
->last_trans_committed
= cur_trans
->transid
;
1419 wake_up(&cur_trans
->commit_wait
);
1421 spin_lock(&root
->fs_info
->trans_lock
);
1422 list_del_init(&cur_trans
->list
);
1423 spin_unlock(&root
->fs_info
->trans_lock
);
1425 put_transaction(cur_trans
);
1426 put_transaction(cur_trans
);
1428 trace_btrfs_transaction_commit(root
);
1430 btrfs_scrub_continue(root
);
1432 if (current
->journal_info
== trans
)
1433 current
->journal_info
= NULL
;
1435 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1437 if (current
!= root
->fs_info
->transaction_kthread
)
1438 btrfs_run_delayed_iputs(root
);
1444 * interface function to delete all the snapshots we have scheduled for deletion
1446 int btrfs_clean_old_snapshots(struct btrfs_root
*root
)
1449 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1451 spin_lock(&fs_info
->trans_lock
);
1452 list_splice_init(&fs_info
->dead_roots
, &list
);
1453 spin_unlock(&fs_info
->trans_lock
);
1455 while (!list_empty(&list
)) {
1456 root
= list_entry(list
.next
, struct btrfs_root
, root_list
);
1457 list_del(&root
->root_list
);
1459 btrfs_kill_all_delayed_nodes(root
);
1461 if (btrfs_header_backref_rev(root
->node
) <
1462 BTRFS_MIXED_BACKREF_REV
)
1463 btrfs_drop_snapshot(root
, NULL
, 0);
1465 btrfs_drop_snapshot(root
, NULL
, 1);