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>
25 #include <linux/uuid.h>
28 #include "transaction.h"
31 #include "inode-map.h"
33 #include "dev-replace.h"
35 #define BTRFS_ROOT_TRANS_TAG 0
37 void put_transaction(struct btrfs_transaction
*transaction
)
39 WARN_ON(atomic_read(&transaction
->use_count
) == 0);
40 if (atomic_dec_and_test(&transaction
->use_count
)) {
41 BUG_ON(!list_empty(&transaction
->list
));
42 WARN_ON(transaction
->delayed_refs
.root
.rb_node
);
43 kmem_cache_free(btrfs_transaction_cachep
, transaction
);
47 static noinline
void switch_commit_root(struct btrfs_root
*root
)
49 free_extent_buffer(root
->commit_root
);
50 root
->commit_root
= btrfs_root_node(root
);
54 * either allocate a new transaction or hop into the existing one
56 static noinline
int join_transaction(struct btrfs_root
*root
, int type
)
58 struct btrfs_transaction
*cur_trans
;
59 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
61 spin_lock(&fs_info
->trans_lock
);
63 /* The file system has been taken offline. No new transactions. */
64 if (fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
) {
65 spin_unlock(&fs_info
->trans_lock
);
69 if (fs_info
->trans_no_join
) {
71 * If we are JOIN_NOLOCK we're already committing a current
72 * transaction, we just need a handle to deal with something
73 * when committing the transaction, such as inode cache and
74 * space cache. It is a special case.
76 if (type
!= TRANS_JOIN_NOLOCK
) {
77 spin_unlock(&fs_info
->trans_lock
);
82 cur_trans
= fs_info
->running_transaction
;
84 if (cur_trans
->aborted
) {
85 spin_unlock(&fs_info
->trans_lock
);
86 return cur_trans
->aborted
;
88 atomic_inc(&cur_trans
->use_count
);
89 atomic_inc(&cur_trans
->num_writers
);
90 cur_trans
->num_joined
++;
91 spin_unlock(&fs_info
->trans_lock
);
94 spin_unlock(&fs_info
->trans_lock
);
97 * If we are ATTACH, we just want to catch the current transaction,
98 * and commit it. If there is no transaction, just return ENOENT.
100 if (type
== TRANS_ATTACH
)
103 cur_trans
= kmem_cache_alloc(btrfs_transaction_cachep
, GFP_NOFS
);
107 spin_lock(&fs_info
->trans_lock
);
108 if (fs_info
->running_transaction
) {
110 * someone started a transaction after we unlocked. Make sure
111 * to redo the trans_no_join checks above
113 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
114 cur_trans
= fs_info
->running_transaction
;
116 } else if (fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
) {
117 spin_unlock(&fs_info
->trans_lock
);
118 kmem_cache_free(btrfs_transaction_cachep
, cur_trans
);
122 atomic_set(&cur_trans
->num_writers
, 1);
123 cur_trans
->num_joined
= 0;
124 init_waitqueue_head(&cur_trans
->writer_wait
);
125 init_waitqueue_head(&cur_trans
->commit_wait
);
126 cur_trans
->in_commit
= 0;
127 cur_trans
->blocked
= 0;
129 * One for this trans handle, one so it will live on until we
130 * commit the transaction.
132 atomic_set(&cur_trans
->use_count
, 2);
133 cur_trans
->commit_done
= 0;
134 cur_trans
->start_time
= get_seconds();
136 cur_trans
->delayed_refs
.root
= RB_ROOT
;
137 cur_trans
->delayed_refs
.num_entries
= 0;
138 cur_trans
->delayed_refs
.num_heads_ready
= 0;
139 cur_trans
->delayed_refs
.num_heads
= 0;
140 cur_trans
->delayed_refs
.flushing
= 0;
141 cur_trans
->delayed_refs
.run_delayed_start
= 0;
144 * although the tree mod log is per file system and not per transaction,
145 * the log must never go across transaction boundaries.
148 if (!list_empty(&fs_info
->tree_mod_seq_list
))
149 WARN(1, KERN_ERR
"btrfs: tree_mod_seq_list not empty when "
150 "creating a fresh transaction\n");
151 if (!RB_EMPTY_ROOT(&fs_info
->tree_mod_log
))
152 WARN(1, KERN_ERR
"btrfs: tree_mod_log rb tree not empty when "
153 "creating a fresh transaction\n");
154 atomic_set(&fs_info
->tree_mod_seq
, 0);
156 spin_lock_init(&cur_trans
->commit_lock
);
157 spin_lock_init(&cur_trans
->delayed_refs
.lock
);
159 INIT_LIST_HEAD(&cur_trans
->pending_snapshots
);
160 list_add_tail(&cur_trans
->list
, &fs_info
->trans_list
);
161 extent_io_tree_init(&cur_trans
->dirty_pages
,
162 fs_info
->btree_inode
->i_mapping
);
163 fs_info
->generation
++;
164 cur_trans
->transid
= fs_info
->generation
;
165 fs_info
->running_transaction
= cur_trans
;
166 cur_trans
->aborted
= 0;
167 spin_unlock(&fs_info
->trans_lock
);
173 * this does all the record keeping required to make sure that a reference
174 * counted root is properly recorded in a given transaction. This is required
175 * to make sure the old root from before we joined the transaction is deleted
176 * when the transaction commits
178 static int record_root_in_trans(struct btrfs_trans_handle
*trans
,
179 struct btrfs_root
*root
)
181 if (root
->ref_cows
&& root
->last_trans
< trans
->transid
) {
182 WARN_ON(root
== root
->fs_info
->extent_root
);
183 WARN_ON(root
->commit_root
!= root
->node
);
186 * see below for in_trans_setup usage rules
187 * we have the reloc mutex held now, so there
188 * is only one writer in this function
190 root
->in_trans_setup
= 1;
192 /* make sure readers find in_trans_setup before
193 * they find our root->last_trans update
197 spin_lock(&root
->fs_info
->fs_roots_radix_lock
);
198 if (root
->last_trans
== trans
->transid
) {
199 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
202 radix_tree_tag_set(&root
->fs_info
->fs_roots_radix
,
203 (unsigned long)root
->root_key
.objectid
,
204 BTRFS_ROOT_TRANS_TAG
);
205 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
206 root
->last_trans
= trans
->transid
;
208 /* this is pretty tricky. We don't want to
209 * take the relocation lock in btrfs_record_root_in_trans
210 * unless we're really doing the first setup for this root in
213 * Normally we'd use root->last_trans as a flag to decide
214 * if we want to take the expensive mutex.
216 * But, we have to set root->last_trans before we
217 * init the relocation root, otherwise, we trip over warnings
218 * in ctree.c. The solution used here is to flag ourselves
219 * with root->in_trans_setup. When this is 1, we're still
220 * fixing up the reloc trees and everyone must wait.
222 * When this is zero, they can trust root->last_trans and fly
223 * through btrfs_record_root_in_trans without having to take the
224 * lock. smp_wmb() makes sure that all the writes above are
225 * done before we pop in the zero below
227 btrfs_init_reloc_root(trans
, root
);
229 root
->in_trans_setup
= 0;
235 int btrfs_record_root_in_trans(struct btrfs_trans_handle
*trans
,
236 struct btrfs_root
*root
)
242 * see record_root_in_trans for comments about in_trans_setup usage
246 if (root
->last_trans
== trans
->transid
&&
247 !root
->in_trans_setup
)
250 mutex_lock(&root
->fs_info
->reloc_mutex
);
251 record_root_in_trans(trans
, root
);
252 mutex_unlock(&root
->fs_info
->reloc_mutex
);
257 /* wait for commit against the current transaction to become unblocked
258 * when this is done, it is safe to start a new transaction, but the current
259 * transaction might not be fully on disk.
261 static void wait_current_trans(struct btrfs_root
*root
)
263 struct btrfs_transaction
*cur_trans
;
265 spin_lock(&root
->fs_info
->trans_lock
);
266 cur_trans
= root
->fs_info
->running_transaction
;
267 if (cur_trans
&& cur_trans
->blocked
) {
268 atomic_inc(&cur_trans
->use_count
);
269 spin_unlock(&root
->fs_info
->trans_lock
);
271 wait_event(root
->fs_info
->transaction_wait
,
272 !cur_trans
->blocked
);
273 put_transaction(cur_trans
);
275 spin_unlock(&root
->fs_info
->trans_lock
);
279 static int may_wait_transaction(struct btrfs_root
*root
, int type
)
281 if (root
->fs_info
->log_root_recovering
)
284 if (type
== TRANS_USERSPACE
)
287 if (type
== TRANS_START
&&
288 !atomic_read(&root
->fs_info
->open_ioctl_trans
))
294 static struct btrfs_trans_handle
*
295 start_transaction(struct btrfs_root
*root
, u64 num_items
, int type
,
296 enum btrfs_reserve_flush_enum flush
)
298 struct btrfs_trans_handle
*h
;
299 struct btrfs_transaction
*cur_trans
;
302 u64 qgroup_reserved
= 0;
304 if (root
->fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
)
305 return ERR_PTR(-EROFS
);
307 if (current
->journal_info
) {
308 WARN_ON(type
!= TRANS_JOIN
&& type
!= TRANS_JOIN_NOLOCK
);
309 h
= current
->journal_info
;
311 WARN_ON(h
->use_count
> 2);
312 h
->orig_rsv
= h
->block_rsv
;
318 * Do the reservation before we join the transaction so we can do all
319 * the appropriate flushing if need be.
321 if (num_items
> 0 && root
!= root
->fs_info
->chunk_root
) {
322 if (root
->fs_info
->quota_enabled
&&
323 is_fstree(root
->root_key
.objectid
)) {
324 qgroup_reserved
= num_items
* root
->leafsize
;
325 ret
= btrfs_qgroup_reserve(root
, qgroup_reserved
);
330 num_bytes
= btrfs_calc_trans_metadata_size(root
, num_items
);
331 ret
= btrfs_block_rsv_add(root
,
332 &root
->fs_info
->trans_block_rsv
,
338 h
= kmem_cache_alloc(btrfs_trans_handle_cachep
, GFP_NOFS
);
345 * If we are JOIN_NOLOCK we're already committing a transaction and
346 * waiting on this guy, so we don't need to do the sb_start_intwrite
347 * because we're already holding a ref. We need this because we could
348 * have raced in and did an fsync() on a file which can kick a commit
349 * and then we deadlock with somebody doing a freeze.
351 * If we are ATTACH, it means we just want to catch the current
352 * transaction and commit it, so we needn't do sb_start_intwrite().
354 if (type
< TRANS_JOIN_NOLOCK
)
355 sb_start_intwrite(root
->fs_info
->sb
);
357 if (may_wait_transaction(root
, type
))
358 wait_current_trans(root
);
361 ret
= join_transaction(root
, type
);
363 wait_current_trans(root
);
364 } while (ret
== -EBUSY
);
367 /* We must get the transaction if we are JOIN_NOLOCK. */
368 BUG_ON(type
== TRANS_JOIN_NOLOCK
);
372 cur_trans
= root
->fs_info
->running_transaction
;
374 h
->transid
= cur_trans
->transid
;
375 h
->transaction
= cur_trans
;
377 h
->bytes_reserved
= 0;
379 h
->delayed_ref_updates
= 0;
385 h
->qgroup_reserved
= qgroup_reserved
;
386 h
->delayed_ref_elem
.seq
= 0;
388 h
->allocating_chunk
= false;
389 INIT_LIST_HEAD(&h
->qgroup_ref_list
);
390 INIT_LIST_HEAD(&h
->new_bgs
);
393 if (cur_trans
->blocked
&& may_wait_transaction(root
, type
)) {
394 btrfs_commit_transaction(h
, root
);
399 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
400 h
->transid
, num_bytes
, 1);
401 h
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
402 h
->bytes_reserved
= num_bytes
;
406 btrfs_record_root_in_trans(h
, root
);
408 if (!current
->journal_info
&& type
!= TRANS_USERSPACE
)
409 current
->journal_info
= h
;
413 if (type
< TRANS_JOIN_NOLOCK
)
414 sb_end_intwrite(root
->fs_info
->sb
);
415 kmem_cache_free(btrfs_trans_handle_cachep
, h
);
418 btrfs_block_rsv_release(root
, &root
->fs_info
->trans_block_rsv
,
422 btrfs_qgroup_free(root
, qgroup_reserved
);
426 struct btrfs_trans_handle
*btrfs_start_transaction(struct btrfs_root
*root
,
429 return start_transaction(root
, num_items
, TRANS_START
,
430 BTRFS_RESERVE_FLUSH_ALL
);
433 struct btrfs_trans_handle
*btrfs_start_transaction_lflush(
434 struct btrfs_root
*root
, int num_items
)
436 return start_transaction(root
, num_items
, TRANS_START
,
437 BTRFS_RESERVE_FLUSH_LIMIT
);
440 struct btrfs_trans_handle
*btrfs_join_transaction(struct btrfs_root
*root
)
442 return start_transaction(root
, 0, TRANS_JOIN
, 0);
445 struct btrfs_trans_handle
*btrfs_join_transaction_nolock(struct btrfs_root
*root
)
447 return start_transaction(root
, 0, TRANS_JOIN_NOLOCK
, 0);
450 struct btrfs_trans_handle
*btrfs_start_ioctl_transaction(struct btrfs_root
*root
)
452 return start_transaction(root
, 0, TRANS_USERSPACE
, 0);
455 struct btrfs_trans_handle
*btrfs_attach_transaction(struct btrfs_root
*root
)
457 return start_transaction(root
, 0, TRANS_ATTACH
, 0);
460 /* wait for a transaction commit to be fully complete */
461 static noinline
void wait_for_commit(struct btrfs_root
*root
,
462 struct btrfs_transaction
*commit
)
464 wait_event(commit
->commit_wait
, commit
->commit_done
);
467 int btrfs_wait_for_commit(struct btrfs_root
*root
, u64 transid
)
469 struct btrfs_transaction
*cur_trans
= NULL
, *t
;
473 if (transid
<= root
->fs_info
->last_trans_committed
)
477 /* find specified transaction */
478 spin_lock(&root
->fs_info
->trans_lock
);
479 list_for_each_entry(t
, &root
->fs_info
->trans_list
, list
) {
480 if (t
->transid
== transid
) {
482 atomic_inc(&cur_trans
->use_count
);
486 if (t
->transid
> transid
) {
491 spin_unlock(&root
->fs_info
->trans_lock
);
492 /* The specified transaction doesn't exist */
496 /* find newest transaction that is committing | committed */
497 spin_lock(&root
->fs_info
->trans_lock
);
498 list_for_each_entry_reverse(t
, &root
->fs_info
->trans_list
,
504 atomic_inc(&cur_trans
->use_count
);
508 spin_unlock(&root
->fs_info
->trans_lock
);
510 goto out
; /* nothing committing|committed */
513 wait_for_commit(root
, cur_trans
);
514 put_transaction(cur_trans
);
519 void btrfs_throttle(struct btrfs_root
*root
)
521 if (!atomic_read(&root
->fs_info
->open_ioctl_trans
))
522 wait_current_trans(root
);
525 static int should_end_transaction(struct btrfs_trans_handle
*trans
,
526 struct btrfs_root
*root
)
530 ret
= btrfs_block_rsv_check(root
, &root
->fs_info
->global_block_rsv
, 5);
534 int btrfs_should_end_transaction(struct btrfs_trans_handle
*trans
,
535 struct btrfs_root
*root
)
537 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
542 if (cur_trans
->blocked
|| cur_trans
->delayed_refs
.flushing
)
545 updates
= trans
->delayed_ref_updates
;
546 trans
->delayed_ref_updates
= 0;
548 err
= btrfs_run_delayed_refs(trans
, root
, updates
);
549 if (err
) /* Error code will also eval true */
553 return should_end_transaction(trans
, root
);
556 static int __btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
557 struct btrfs_root
*root
, int throttle
)
559 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
560 struct btrfs_fs_info
*info
= root
->fs_info
;
562 int lock
= (trans
->type
!= TRANS_JOIN_NOLOCK
);
565 if (--trans
->use_count
) {
566 trans
->block_rsv
= trans
->orig_rsv
;
571 * do the qgroup accounting as early as possible
573 err
= btrfs_delayed_refs_qgroup_accounting(trans
, info
);
575 btrfs_trans_release_metadata(trans
, root
);
576 trans
->block_rsv
= NULL
;
578 * the same root has to be passed to start_transaction and
579 * end_transaction. Subvolume quota depends on this.
581 WARN_ON(trans
->root
!= root
);
583 if (trans
->qgroup_reserved
) {
584 btrfs_qgroup_free(root
, trans
->qgroup_reserved
);
585 trans
->qgroup_reserved
= 0;
588 if (!list_empty(&trans
->new_bgs
))
589 btrfs_create_pending_block_groups(trans
, root
);
592 unsigned long cur
= trans
->delayed_ref_updates
;
593 trans
->delayed_ref_updates
= 0;
595 trans
->transaction
->delayed_refs
.num_heads_ready
> 64) {
596 trans
->delayed_ref_updates
= 0;
597 btrfs_run_delayed_refs(trans
, root
, cur
);
603 btrfs_trans_release_metadata(trans
, root
);
604 trans
->block_rsv
= NULL
;
606 if (!list_empty(&trans
->new_bgs
))
607 btrfs_create_pending_block_groups(trans
, root
);
609 if (lock
&& !atomic_read(&root
->fs_info
->open_ioctl_trans
) &&
610 should_end_transaction(trans
, root
)) {
611 trans
->transaction
->blocked
= 1;
615 if (lock
&& cur_trans
->blocked
&& !cur_trans
->in_commit
) {
618 * We may race with somebody else here so end up having
619 * to call end_transaction on ourselves again, so inc
623 return btrfs_commit_transaction(trans
, root
);
625 wake_up_process(info
->transaction_kthread
);
629 if (trans
->type
< TRANS_JOIN_NOLOCK
)
630 sb_end_intwrite(root
->fs_info
->sb
);
632 WARN_ON(cur_trans
!= info
->running_transaction
);
633 WARN_ON(atomic_read(&cur_trans
->num_writers
) < 1);
634 atomic_dec(&cur_trans
->num_writers
);
637 if (waitqueue_active(&cur_trans
->writer_wait
))
638 wake_up(&cur_trans
->writer_wait
);
639 put_transaction(cur_trans
);
641 if (current
->journal_info
== trans
)
642 current
->journal_info
= NULL
;
645 btrfs_run_delayed_iputs(root
);
647 if (trans
->aborted
||
648 root
->fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
) {
651 assert_qgroups_uptodate(trans
);
653 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
657 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
658 struct btrfs_root
*root
)
662 ret
= __btrfs_end_transaction(trans
, root
, 0);
668 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
,
669 struct btrfs_root
*root
)
673 ret
= __btrfs_end_transaction(trans
, root
, 1);
679 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle
*trans
,
680 struct btrfs_root
*root
)
682 return __btrfs_end_transaction(trans
, root
, 1);
686 * when btree blocks are allocated, they have some corresponding bits set for
687 * them in one of two extent_io trees. This is used to make sure all of
688 * those extents are sent to disk but does not wait on them
690 int btrfs_write_marked_extents(struct btrfs_root
*root
,
691 struct extent_io_tree
*dirty_pages
, int mark
)
695 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
696 struct extent_state
*cached_state
= NULL
;
700 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
701 mark
, &cached_state
)) {
702 convert_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
,
703 mark
, &cached_state
, GFP_NOFS
);
705 err
= filemap_fdatawrite_range(mapping
, start
, end
);
717 * when btree blocks are allocated, they have some corresponding bits set for
718 * them in one of two extent_io trees. This is used to make sure all of
719 * those extents are on disk for transaction or log commit. We wait
720 * on all the pages and clear them from the dirty pages state tree
722 int btrfs_wait_marked_extents(struct btrfs_root
*root
,
723 struct extent_io_tree
*dirty_pages
, int mark
)
727 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
728 struct extent_state
*cached_state
= NULL
;
732 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
733 EXTENT_NEED_WAIT
, &cached_state
)) {
734 clear_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
,
735 0, 0, &cached_state
, GFP_NOFS
);
736 err
= filemap_fdatawait_range(mapping
, start
, end
);
748 * when btree blocks are allocated, they have some corresponding bits set for
749 * them in one of two extent_io trees. This is used to make sure all of
750 * those extents are on disk for transaction or log commit
752 int btrfs_write_and_wait_marked_extents(struct btrfs_root
*root
,
753 struct extent_io_tree
*dirty_pages
, int mark
)
758 ret
= btrfs_write_marked_extents(root
, dirty_pages
, mark
);
759 ret2
= btrfs_wait_marked_extents(root
, dirty_pages
, mark
);
768 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
,
769 struct btrfs_root
*root
)
771 if (!trans
|| !trans
->transaction
) {
772 struct inode
*btree_inode
;
773 btree_inode
= root
->fs_info
->btree_inode
;
774 return filemap_write_and_wait(btree_inode
->i_mapping
);
776 return btrfs_write_and_wait_marked_extents(root
,
777 &trans
->transaction
->dirty_pages
,
782 * this is used to update the root pointer in the tree of tree roots.
784 * But, in the case of the extent allocation tree, updating the root
785 * pointer may allocate blocks which may change the root of the extent
788 * So, this loops and repeats and makes sure the cowonly root didn't
789 * change while the root pointer was being updated in the metadata.
791 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
792 struct btrfs_root
*root
)
797 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
799 old_root_used
= btrfs_root_used(&root
->root_item
);
800 btrfs_write_dirty_block_groups(trans
, root
);
803 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
804 if (old_root_bytenr
== root
->node
->start
&&
805 old_root_used
== btrfs_root_used(&root
->root_item
))
808 btrfs_set_root_node(&root
->root_item
, root
->node
);
809 ret
= btrfs_update_root(trans
, tree_root
,
815 old_root_used
= btrfs_root_used(&root
->root_item
);
816 ret
= btrfs_write_dirty_block_groups(trans
, root
);
821 if (root
!= root
->fs_info
->extent_root
)
822 switch_commit_root(root
);
828 * update all the cowonly tree roots on disk
830 * The error handling in this function may not be obvious. Any of the
831 * failures will cause the file system to go offline. We still need
832 * to clean up the delayed refs.
834 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
,
835 struct btrfs_root
*root
)
837 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
838 struct list_head
*next
;
839 struct extent_buffer
*eb
;
842 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
846 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
847 ret
= btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
,
849 btrfs_tree_unlock(eb
);
850 free_extent_buffer(eb
);
855 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
859 ret
= btrfs_run_dev_stats(trans
, root
->fs_info
);
861 ret
= btrfs_run_dev_replace(trans
, root
->fs_info
);
864 ret
= btrfs_run_qgroups(trans
, root
->fs_info
);
867 /* run_qgroups might have added some more refs */
868 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
871 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
872 next
= fs_info
->dirty_cowonly_roots
.next
;
874 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
876 ret
= update_cowonly_root(trans
, root
);
881 down_write(&fs_info
->extent_commit_sem
);
882 switch_commit_root(fs_info
->extent_root
);
883 up_write(&fs_info
->extent_commit_sem
);
885 btrfs_after_dev_replace_commit(fs_info
);
891 * dead roots are old snapshots that need to be deleted. This allocates
892 * a dirty root struct and adds it into the list of dead roots that need to
895 int btrfs_add_dead_root(struct btrfs_root
*root
)
897 spin_lock(&root
->fs_info
->trans_lock
);
898 list_add(&root
->root_list
, &root
->fs_info
->dead_roots
);
899 spin_unlock(&root
->fs_info
->trans_lock
);
904 * update all the cowonly tree roots on disk
906 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
,
907 struct btrfs_root
*root
)
909 struct btrfs_root
*gang
[8];
910 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
915 spin_lock(&fs_info
->fs_roots_radix_lock
);
917 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
920 BTRFS_ROOT_TRANS_TAG
);
923 for (i
= 0; i
< ret
; i
++) {
925 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
926 (unsigned long)root
->root_key
.objectid
,
927 BTRFS_ROOT_TRANS_TAG
);
928 spin_unlock(&fs_info
->fs_roots_radix_lock
);
930 btrfs_free_log(trans
, root
);
931 btrfs_update_reloc_root(trans
, root
);
932 btrfs_orphan_commit_root(trans
, root
);
934 btrfs_save_ino_cache(root
, trans
);
936 /* see comments in should_cow_block() */
940 if (root
->commit_root
!= root
->node
) {
941 mutex_lock(&root
->fs_commit_mutex
);
942 switch_commit_root(root
);
943 btrfs_unpin_free_ino(root
);
944 mutex_unlock(&root
->fs_commit_mutex
);
946 btrfs_set_root_node(&root
->root_item
,
950 err
= btrfs_update_root(trans
, fs_info
->tree_root
,
953 spin_lock(&fs_info
->fs_roots_radix_lock
);
958 spin_unlock(&fs_info
->fs_roots_radix_lock
);
963 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
964 * otherwise every leaf in the btree is read and defragged.
966 int btrfs_defrag_root(struct btrfs_root
*root
, int cacheonly
)
968 struct btrfs_fs_info
*info
= root
->fs_info
;
969 struct btrfs_trans_handle
*trans
;
972 if (xchg(&root
->defrag_running
, 1))
976 trans
= btrfs_start_transaction(root
, 0);
978 return PTR_ERR(trans
);
980 ret
= btrfs_defrag_leaves(trans
, root
, cacheonly
);
982 btrfs_end_transaction(trans
, root
);
983 btrfs_btree_balance_dirty(info
->tree_root
);
986 if (btrfs_fs_closing(root
->fs_info
) || ret
!= -EAGAIN
)
989 root
->defrag_running
= 0;
994 * new snapshots need to be created at a very specific time in the
995 * transaction commit. This does the actual creation
997 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
998 struct btrfs_fs_info
*fs_info
,
999 struct btrfs_pending_snapshot
*pending
)
1001 struct btrfs_key key
;
1002 struct btrfs_root_item
*new_root_item
;
1003 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1004 struct btrfs_root
*root
= pending
->root
;
1005 struct btrfs_root
*parent_root
;
1006 struct btrfs_block_rsv
*rsv
;
1007 struct inode
*parent_inode
;
1008 struct btrfs_path
*path
;
1009 struct btrfs_dir_item
*dir_item
;
1010 struct dentry
*parent
;
1011 struct dentry
*dentry
;
1012 struct extent_buffer
*tmp
;
1013 struct extent_buffer
*old
;
1014 struct timespec cur_time
= CURRENT_TIME
;
1022 path
= btrfs_alloc_path();
1024 ret
= pending
->error
= -ENOMEM
;
1025 goto path_alloc_fail
;
1028 new_root_item
= kmalloc(sizeof(*new_root_item
), GFP_NOFS
);
1029 if (!new_root_item
) {
1030 ret
= pending
->error
= -ENOMEM
;
1031 goto root_item_alloc_fail
;
1034 ret
= btrfs_find_free_objectid(tree_root
, &objectid
);
1036 pending
->error
= ret
;
1037 goto no_free_objectid
;
1040 btrfs_reloc_pre_snapshot(trans
, pending
, &to_reserve
);
1042 if (to_reserve
> 0) {
1043 ret
= btrfs_block_rsv_add(root
, &pending
->block_rsv
,
1045 BTRFS_RESERVE_NO_FLUSH
);
1047 pending
->error
= ret
;
1048 goto no_free_objectid
;
1052 ret
= btrfs_qgroup_inherit(trans
, fs_info
, root
->root_key
.objectid
,
1053 objectid
, pending
->inherit
);
1055 pending
->error
= ret
;
1056 goto no_free_objectid
;
1059 key
.objectid
= objectid
;
1060 key
.offset
= (u64
)-1;
1061 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1063 rsv
= trans
->block_rsv
;
1064 trans
->block_rsv
= &pending
->block_rsv
;
1066 dentry
= pending
->dentry
;
1067 parent
= dget_parent(dentry
);
1068 parent_inode
= parent
->d_inode
;
1069 parent_root
= BTRFS_I(parent_inode
)->root
;
1070 record_root_in_trans(trans
, parent_root
);
1073 * insert the directory item
1075 ret
= btrfs_set_inode_index(parent_inode
, &index
);
1076 BUG_ON(ret
); /* -ENOMEM */
1078 /* check if there is a file/dir which has the same name. */
1079 dir_item
= btrfs_lookup_dir_item(NULL
, parent_root
, path
,
1080 btrfs_ino(parent_inode
),
1081 dentry
->d_name
.name
,
1082 dentry
->d_name
.len
, 0);
1083 if (dir_item
!= NULL
&& !IS_ERR(dir_item
)) {
1084 pending
->error
= -EEXIST
;
1086 } else if (IS_ERR(dir_item
)) {
1087 ret
= PTR_ERR(dir_item
);
1088 btrfs_abort_transaction(trans
, root
, ret
);
1091 btrfs_release_path(path
);
1094 * pull in the delayed directory update
1095 * and the delayed inode item
1096 * otherwise we corrupt the FS during
1099 ret
= btrfs_run_delayed_items(trans
, root
);
1100 if (ret
) { /* Transaction aborted */
1101 btrfs_abort_transaction(trans
, root
, ret
);
1105 record_root_in_trans(trans
, root
);
1106 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
1107 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
1108 btrfs_check_and_init_root_item(new_root_item
);
1110 root_flags
= btrfs_root_flags(new_root_item
);
1111 if (pending
->readonly
)
1112 root_flags
|= BTRFS_ROOT_SUBVOL_RDONLY
;
1114 root_flags
&= ~BTRFS_ROOT_SUBVOL_RDONLY
;
1115 btrfs_set_root_flags(new_root_item
, root_flags
);
1117 btrfs_set_root_generation_v2(new_root_item
,
1119 uuid_le_gen(&new_uuid
);
1120 memcpy(new_root_item
->uuid
, new_uuid
.b
, BTRFS_UUID_SIZE
);
1121 memcpy(new_root_item
->parent_uuid
, root
->root_item
.uuid
,
1123 new_root_item
->otime
.sec
= cpu_to_le64(cur_time
.tv_sec
);
1124 new_root_item
->otime
.nsec
= cpu_to_le32(cur_time
.tv_nsec
);
1125 btrfs_set_root_otransid(new_root_item
, trans
->transid
);
1126 memset(&new_root_item
->stime
, 0, sizeof(new_root_item
->stime
));
1127 memset(&new_root_item
->rtime
, 0, sizeof(new_root_item
->rtime
));
1128 btrfs_set_root_stransid(new_root_item
, 0);
1129 btrfs_set_root_rtransid(new_root_item
, 0);
1131 old
= btrfs_lock_root_node(root
);
1132 ret
= btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
);
1134 btrfs_tree_unlock(old
);
1135 free_extent_buffer(old
);
1136 btrfs_abort_transaction(trans
, root
, ret
);
1140 btrfs_set_lock_blocking(old
);
1142 ret
= btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
1143 /* clean up in any case */
1144 btrfs_tree_unlock(old
);
1145 free_extent_buffer(old
);
1147 btrfs_abort_transaction(trans
, root
, ret
);
1151 /* see comments in should_cow_block() */
1152 root
->force_cow
= 1;
1155 btrfs_set_root_node(new_root_item
, tmp
);
1156 /* record when the snapshot was created in key.offset */
1157 key
.offset
= trans
->transid
;
1158 ret
= btrfs_insert_root(trans
, tree_root
, &key
, new_root_item
);
1159 btrfs_tree_unlock(tmp
);
1160 free_extent_buffer(tmp
);
1162 btrfs_abort_transaction(trans
, root
, ret
);
1167 * insert root back/forward references
1169 ret
= btrfs_add_root_ref(trans
, tree_root
, objectid
,
1170 parent_root
->root_key
.objectid
,
1171 btrfs_ino(parent_inode
), index
,
1172 dentry
->d_name
.name
, dentry
->d_name
.len
);
1174 btrfs_abort_transaction(trans
, root
, ret
);
1178 key
.offset
= (u64
)-1;
1179 pending
->snap
= btrfs_read_fs_root_no_name(root
->fs_info
, &key
);
1180 if (IS_ERR(pending
->snap
)) {
1181 ret
= PTR_ERR(pending
->snap
);
1182 btrfs_abort_transaction(trans
, root
, ret
);
1186 ret
= btrfs_reloc_post_snapshot(trans
, pending
);
1188 btrfs_abort_transaction(trans
, root
, ret
);
1192 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1194 btrfs_abort_transaction(trans
, root
, ret
);
1198 ret
= btrfs_insert_dir_item(trans
, parent_root
,
1199 dentry
->d_name
.name
, dentry
->d_name
.len
,
1201 BTRFS_FT_DIR
, index
);
1202 /* We have check then name at the beginning, so it is impossible. */
1203 BUG_ON(ret
== -EEXIST
|| ret
== -EOVERFLOW
);
1205 btrfs_abort_transaction(trans
, root
, ret
);
1209 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
1210 dentry
->d_name
.len
* 2);
1211 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
1212 ret
= btrfs_update_inode_fallback(trans
, parent_root
, parent_inode
);
1214 btrfs_abort_transaction(trans
, root
, ret
);
1217 trans
->block_rsv
= rsv
;
1219 kfree(new_root_item
);
1220 root_item_alloc_fail
:
1221 btrfs_free_path(path
);
1223 btrfs_block_rsv_release(root
, &pending
->block_rsv
, (u64
)-1);
1228 * create all the snapshots we've scheduled for creation
1230 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
,
1231 struct btrfs_fs_info
*fs_info
)
1233 struct btrfs_pending_snapshot
*pending
;
1234 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
1236 list_for_each_entry(pending
, head
, list
)
1237 create_pending_snapshot(trans
, fs_info
, pending
);
1241 static void update_super_roots(struct btrfs_root
*root
)
1243 struct btrfs_root_item
*root_item
;
1244 struct btrfs_super_block
*super
;
1246 super
= root
->fs_info
->super_copy
;
1248 root_item
= &root
->fs_info
->chunk_root
->root_item
;
1249 super
->chunk_root
= root_item
->bytenr
;
1250 super
->chunk_root_generation
= root_item
->generation
;
1251 super
->chunk_root_level
= root_item
->level
;
1253 root_item
= &root
->fs_info
->tree_root
->root_item
;
1254 super
->root
= root_item
->bytenr
;
1255 super
->generation
= root_item
->generation
;
1256 super
->root_level
= root_item
->level
;
1257 if (btrfs_test_opt(root
, SPACE_CACHE
))
1258 super
->cache_generation
= root_item
->generation
;
1261 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
1264 spin_lock(&info
->trans_lock
);
1265 if (info
->running_transaction
)
1266 ret
= info
->running_transaction
->in_commit
;
1267 spin_unlock(&info
->trans_lock
);
1271 int btrfs_transaction_blocked(struct btrfs_fs_info
*info
)
1274 spin_lock(&info
->trans_lock
);
1275 if (info
->running_transaction
)
1276 ret
= info
->running_transaction
->blocked
;
1277 spin_unlock(&info
->trans_lock
);
1282 * wait for the current transaction commit to start and block subsequent
1285 static void wait_current_trans_commit_start(struct btrfs_root
*root
,
1286 struct btrfs_transaction
*trans
)
1288 wait_event(root
->fs_info
->transaction_blocked_wait
, trans
->in_commit
);
1292 * wait for the current transaction to start and then become unblocked.
1295 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root
*root
,
1296 struct btrfs_transaction
*trans
)
1298 wait_event(root
->fs_info
->transaction_wait
,
1299 trans
->commit_done
|| (trans
->in_commit
&& !trans
->blocked
));
1303 * commit transactions asynchronously. once btrfs_commit_transaction_async
1304 * returns, any subsequent transaction will not be allowed to join.
1306 struct btrfs_async_commit
{
1307 struct btrfs_trans_handle
*newtrans
;
1308 struct btrfs_root
*root
;
1309 struct work_struct work
;
1312 static void do_async_commit(struct work_struct
*work
)
1314 struct btrfs_async_commit
*ac
=
1315 container_of(work
, struct btrfs_async_commit
, work
);
1318 * We've got freeze protection passed with the transaction.
1319 * Tell lockdep about it.
1321 if (ac
->newtrans
->type
< TRANS_JOIN_NOLOCK
)
1323 &ac
->root
->fs_info
->sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
1326 current
->journal_info
= ac
->newtrans
;
1328 btrfs_commit_transaction(ac
->newtrans
, ac
->root
);
1332 int btrfs_commit_transaction_async(struct btrfs_trans_handle
*trans
,
1333 struct btrfs_root
*root
,
1334 int wait_for_unblock
)
1336 struct btrfs_async_commit
*ac
;
1337 struct btrfs_transaction
*cur_trans
;
1339 ac
= kmalloc(sizeof(*ac
), GFP_NOFS
);
1343 INIT_WORK(&ac
->work
, do_async_commit
);
1345 ac
->newtrans
= btrfs_join_transaction(root
);
1346 if (IS_ERR(ac
->newtrans
)) {
1347 int err
= PTR_ERR(ac
->newtrans
);
1352 /* take transaction reference */
1353 cur_trans
= trans
->transaction
;
1354 atomic_inc(&cur_trans
->use_count
);
1356 btrfs_end_transaction(trans
, root
);
1359 * Tell lockdep we've released the freeze rwsem, since the
1360 * async commit thread will be the one to unlock it.
1362 if (trans
->type
< TRANS_JOIN_NOLOCK
)
1364 &root
->fs_info
->sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
1367 schedule_work(&ac
->work
);
1369 /* wait for transaction to start and unblock */
1370 if (wait_for_unblock
)
1371 wait_current_trans_commit_start_and_unblock(root
, cur_trans
);
1373 wait_current_trans_commit_start(root
, cur_trans
);
1375 if (current
->journal_info
== trans
)
1376 current
->journal_info
= NULL
;
1378 put_transaction(cur_trans
);
1383 static void cleanup_transaction(struct btrfs_trans_handle
*trans
,
1384 struct btrfs_root
*root
, int err
)
1386 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1388 WARN_ON(trans
->use_count
> 1);
1390 btrfs_abort_transaction(trans
, root
, err
);
1392 spin_lock(&root
->fs_info
->trans_lock
);
1393 list_del_init(&cur_trans
->list
);
1394 if (cur_trans
== root
->fs_info
->running_transaction
) {
1395 root
->fs_info
->running_transaction
= NULL
;
1396 root
->fs_info
->trans_no_join
= 0;
1398 spin_unlock(&root
->fs_info
->trans_lock
);
1400 btrfs_cleanup_one_transaction(trans
->transaction
, root
);
1402 put_transaction(cur_trans
);
1403 put_transaction(cur_trans
);
1405 trace_btrfs_transaction_commit(root
);
1407 btrfs_scrub_continue(root
);
1409 if (current
->journal_info
== trans
)
1410 current
->journal_info
= NULL
;
1412 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1415 static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle
*trans
,
1416 struct btrfs_root
*root
)
1418 int flush_on_commit
= btrfs_test_opt(root
, FLUSHONCOMMIT
);
1419 int snap_pending
= 0;
1422 if (!flush_on_commit
) {
1423 spin_lock(&root
->fs_info
->trans_lock
);
1424 if (!list_empty(&trans
->transaction
->pending_snapshots
))
1426 spin_unlock(&root
->fs_info
->trans_lock
);
1429 if (flush_on_commit
|| snap_pending
) {
1430 ret
= btrfs_start_delalloc_inodes(root
, 1);
1433 btrfs_wait_ordered_extents(root
, 1);
1436 ret
= btrfs_run_delayed_items(trans
, root
);
1441 * running the delayed items may have added new refs. account
1442 * them now so that they hinder processing of more delayed refs
1443 * as little as possible.
1445 btrfs_delayed_refs_qgroup_accounting(trans
, root
->fs_info
);
1448 * rename don't use btrfs_join_transaction, so, once we
1449 * set the transaction to blocked above, we aren't going
1450 * to get any new ordered operations. We can safely run
1451 * it here and no for sure that nothing new will be added
1454 btrfs_run_ordered_operations(root
, 1);
1460 * btrfs_transaction state sequence:
1461 * in_commit = 0, blocked = 0 (initial)
1462 * in_commit = 1, blocked = 1
1466 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
,
1467 struct btrfs_root
*root
)
1469 unsigned long joined
= 0;
1470 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1471 struct btrfs_transaction
*prev_trans
= NULL
;
1474 int should_grow
= 0;
1475 unsigned long now
= get_seconds();
1477 ret
= btrfs_run_ordered_operations(root
, 0);
1479 btrfs_abort_transaction(trans
, root
, ret
);
1480 goto cleanup_transaction
;
1483 /* Stop the commit early if ->aborted is set */
1484 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1485 ret
= cur_trans
->aborted
;
1486 goto cleanup_transaction
;
1489 /* make a pass through all the delayed refs we have so far
1490 * any runnings procs may add more while we are here
1492 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1494 goto cleanup_transaction
;
1496 btrfs_trans_release_metadata(trans
, root
);
1497 trans
->block_rsv
= NULL
;
1499 cur_trans
= trans
->transaction
;
1502 * set the flushing flag so procs in this transaction have to
1503 * start sending their work down.
1505 cur_trans
->delayed_refs
.flushing
= 1;
1507 if (!list_empty(&trans
->new_bgs
))
1508 btrfs_create_pending_block_groups(trans
, root
);
1510 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1512 goto cleanup_transaction
;
1514 spin_lock(&cur_trans
->commit_lock
);
1515 if (cur_trans
->in_commit
) {
1516 spin_unlock(&cur_trans
->commit_lock
);
1517 atomic_inc(&cur_trans
->use_count
);
1518 ret
= btrfs_end_transaction(trans
, root
);
1520 wait_for_commit(root
, cur_trans
);
1522 put_transaction(cur_trans
);
1527 trans
->transaction
->in_commit
= 1;
1528 trans
->transaction
->blocked
= 1;
1529 spin_unlock(&cur_trans
->commit_lock
);
1530 wake_up(&root
->fs_info
->transaction_blocked_wait
);
1532 spin_lock(&root
->fs_info
->trans_lock
);
1533 if (cur_trans
->list
.prev
!= &root
->fs_info
->trans_list
) {
1534 prev_trans
= list_entry(cur_trans
->list
.prev
,
1535 struct btrfs_transaction
, list
);
1536 if (!prev_trans
->commit_done
) {
1537 atomic_inc(&prev_trans
->use_count
);
1538 spin_unlock(&root
->fs_info
->trans_lock
);
1540 wait_for_commit(root
, prev_trans
);
1542 put_transaction(prev_trans
);
1544 spin_unlock(&root
->fs_info
->trans_lock
);
1547 spin_unlock(&root
->fs_info
->trans_lock
);
1550 if (!btrfs_test_opt(root
, SSD
) &&
1551 (now
< cur_trans
->start_time
|| now
- cur_trans
->start_time
< 1))
1555 joined
= cur_trans
->num_joined
;
1557 WARN_ON(cur_trans
!= trans
->transaction
);
1559 ret
= btrfs_flush_all_pending_stuffs(trans
, root
);
1561 goto cleanup_transaction
;
1563 prepare_to_wait(&cur_trans
->writer_wait
, &wait
,
1564 TASK_UNINTERRUPTIBLE
);
1566 if (atomic_read(&cur_trans
->num_writers
) > 1)
1567 schedule_timeout(MAX_SCHEDULE_TIMEOUT
);
1568 else if (should_grow
)
1569 schedule_timeout(1);
1571 finish_wait(&cur_trans
->writer_wait
, &wait
);
1572 } while (atomic_read(&cur_trans
->num_writers
) > 1 ||
1573 (should_grow
&& cur_trans
->num_joined
!= joined
));
1575 ret
= btrfs_flush_all_pending_stuffs(trans
, root
);
1577 goto cleanup_transaction
;
1580 * Ok now we need to make sure to block out any other joins while we
1581 * commit the transaction. We could have started a join before setting
1582 * no_join so make sure to wait for num_writers to == 1 again.
1584 spin_lock(&root
->fs_info
->trans_lock
);
1585 root
->fs_info
->trans_no_join
= 1;
1586 spin_unlock(&root
->fs_info
->trans_lock
);
1587 wait_event(cur_trans
->writer_wait
,
1588 atomic_read(&cur_trans
->num_writers
) == 1);
1590 /* ->aborted might be set after the previous check, so check it */
1591 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1592 ret
= cur_trans
->aborted
;
1593 goto cleanup_transaction
;
1596 * the reloc mutex makes sure that we stop
1597 * the balancing code from coming in and moving
1598 * extents around in the middle of the commit
1600 mutex_lock(&root
->fs_info
->reloc_mutex
);
1603 * We needn't worry about the delayed items because we will
1604 * deal with them in create_pending_snapshot(), which is the
1605 * core function of the snapshot creation.
1607 ret
= create_pending_snapshots(trans
, root
->fs_info
);
1609 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1610 goto cleanup_transaction
;
1614 * We insert the dir indexes of the snapshots and update the inode
1615 * of the snapshots' parents after the snapshot creation, so there
1616 * are some delayed items which are not dealt with. Now deal with
1619 * We needn't worry that this operation will corrupt the snapshots,
1620 * because all the tree which are snapshoted will be forced to COW
1621 * the nodes and leaves.
1623 ret
= btrfs_run_delayed_items(trans
, root
);
1625 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1626 goto cleanup_transaction
;
1629 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1631 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1632 goto cleanup_transaction
;
1636 * make sure none of the code above managed to slip in a
1639 btrfs_assert_delayed_root_empty(root
);
1641 WARN_ON(cur_trans
!= trans
->transaction
);
1643 btrfs_scrub_pause(root
);
1644 /* btrfs_commit_tree_roots is responsible for getting the
1645 * various roots consistent with each other. Every pointer
1646 * in the tree of tree roots has to point to the most up to date
1647 * root for every subvolume and other tree. So, we have to keep
1648 * the tree logging code from jumping in and changing any
1651 * At this point in the commit, there can't be any tree-log
1652 * writers, but a little lower down we drop the trans mutex
1653 * and let new people in. By holding the tree_log_mutex
1654 * from now until after the super is written, we avoid races
1655 * with the tree-log code.
1657 mutex_lock(&root
->fs_info
->tree_log_mutex
);
1659 ret
= commit_fs_roots(trans
, root
);
1661 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1662 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1663 goto cleanup_transaction
;
1666 /* commit_fs_roots gets rid of all the tree log roots, it is now
1667 * safe to free the root of tree log roots
1669 btrfs_free_log_root_tree(trans
, root
->fs_info
);
1671 ret
= commit_cowonly_roots(trans
, root
);
1673 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1674 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1675 goto cleanup_transaction
;
1679 * The tasks which save the space cache and inode cache may also
1680 * update ->aborted, check it.
1682 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1683 ret
= cur_trans
->aborted
;
1684 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1685 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1686 goto cleanup_transaction
;
1689 btrfs_prepare_extent_commit(trans
, root
);
1691 cur_trans
= root
->fs_info
->running_transaction
;
1693 btrfs_set_root_node(&root
->fs_info
->tree_root
->root_item
,
1694 root
->fs_info
->tree_root
->node
);
1695 switch_commit_root(root
->fs_info
->tree_root
);
1697 btrfs_set_root_node(&root
->fs_info
->chunk_root
->root_item
,
1698 root
->fs_info
->chunk_root
->node
);
1699 switch_commit_root(root
->fs_info
->chunk_root
);
1701 assert_qgroups_uptodate(trans
);
1702 update_super_roots(root
);
1704 if (!root
->fs_info
->log_root_recovering
) {
1705 btrfs_set_super_log_root(root
->fs_info
->super_copy
, 0);
1706 btrfs_set_super_log_root_level(root
->fs_info
->super_copy
, 0);
1709 memcpy(root
->fs_info
->super_for_commit
, root
->fs_info
->super_copy
,
1710 sizeof(*root
->fs_info
->super_copy
));
1712 trans
->transaction
->blocked
= 0;
1713 spin_lock(&root
->fs_info
->trans_lock
);
1714 root
->fs_info
->running_transaction
= NULL
;
1715 root
->fs_info
->trans_no_join
= 0;
1716 spin_unlock(&root
->fs_info
->trans_lock
);
1717 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1719 wake_up(&root
->fs_info
->transaction_wait
);
1721 ret
= btrfs_write_and_wait_transaction(trans
, root
);
1723 btrfs_error(root
->fs_info
, ret
,
1724 "Error while writing out transaction.");
1725 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1726 goto cleanup_transaction
;
1729 ret
= write_ctree_super(trans
, root
, 0);
1731 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1732 goto cleanup_transaction
;
1736 * the super is written, we can safely allow the tree-loggers
1737 * to go about their business
1739 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1741 btrfs_finish_extent_commit(trans
, root
);
1743 cur_trans
->commit_done
= 1;
1745 root
->fs_info
->last_trans_committed
= cur_trans
->transid
;
1747 wake_up(&cur_trans
->commit_wait
);
1749 spin_lock(&root
->fs_info
->trans_lock
);
1750 list_del_init(&cur_trans
->list
);
1751 spin_unlock(&root
->fs_info
->trans_lock
);
1753 put_transaction(cur_trans
);
1754 put_transaction(cur_trans
);
1756 if (trans
->type
< TRANS_JOIN_NOLOCK
)
1757 sb_end_intwrite(root
->fs_info
->sb
);
1759 trace_btrfs_transaction_commit(root
);
1761 btrfs_scrub_continue(root
);
1763 if (current
->journal_info
== trans
)
1764 current
->journal_info
= NULL
;
1766 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1768 if (current
!= root
->fs_info
->transaction_kthread
)
1769 btrfs_run_delayed_iputs(root
);
1773 cleanup_transaction
:
1774 btrfs_trans_release_metadata(trans
, root
);
1775 trans
->block_rsv
= NULL
;
1776 btrfs_printk(root
->fs_info
, "Skipping commit of aborted transaction.\n");
1778 if (current
->journal_info
== trans
)
1779 current
->journal_info
= NULL
;
1780 cleanup_transaction(trans
, root
, ret
);
1786 * interface function to delete all the snapshots we have scheduled for deletion
1788 int btrfs_clean_old_snapshots(struct btrfs_root
*root
)
1791 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1793 spin_lock(&fs_info
->trans_lock
);
1794 list_splice_init(&fs_info
->dead_roots
, &list
);
1795 spin_unlock(&fs_info
->trans_lock
);
1797 while (!list_empty(&list
)) {
1800 root
= list_entry(list
.next
, struct btrfs_root
, root_list
);
1801 list_del(&root
->root_list
);
1803 btrfs_kill_all_delayed_nodes(root
);
1805 if (btrfs_header_backref_rev(root
->node
) <
1806 BTRFS_MIXED_BACKREF_REV
)
1807 ret
= btrfs_drop_snapshot(root
, NULL
, 0, 0);
1809 ret
=btrfs_drop_snapshot(root
, NULL
, 1, 0);