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 (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
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 (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
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 (test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
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 test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
650 assert_qgroups_uptodate(trans
);
652 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
656 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
657 struct btrfs_root
*root
)
661 ret
= __btrfs_end_transaction(trans
, root
, 0);
667 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
,
668 struct btrfs_root
*root
)
672 ret
= __btrfs_end_transaction(trans
, root
, 1);
678 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle
*trans
,
679 struct btrfs_root
*root
)
681 return __btrfs_end_transaction(trans
, root
, 1);
685 * when btree blocks are allocated, they have some corresponding bits set for
686 * them in one of two extent_io trees. This is used to make sure all of
687 * those extents are sent to disk but does not wait on them
689 int btrfs_write_marked_extents(struct btrfs_root
*root
,
690 struct extent_io_tree
*dirty_pages
, int mark
)
694 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
695 struct extent_state
*cached_state
= NULL
;
699 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
700 mark
, &cached_state
)) {
701 convert_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
,
702 mark
, &cached_state
, GFP_NOFS
);
704 err
= filemap_fdatawrite_range(mapping
, start
, end
);
716 * when btree blocks are allocated, they have some corresponding bits set for
717 * them in one of two extent_io trees. This is used to make sure all of
718 * those extents are on disk for transaction or log commit. We wait
719 * on all the pages and clear them from the dirty pages state tree
721 int btrfs_wait_marked_extents(struct btrfs_root
*root
,
722 struct extent_io_tree
*dirty_pages
, int mark
)
726 struct address_space
*mapping
= root
->fs_info
->btree_inode
->i_mapping
;
727 struct extent_state
*cached_state
= NULL
;
731 while (!find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
732 EXTENT_NEED_WAIT
, &cached_state
)) {
733 clear_extent_bit(dirty_pages
, start
, end
, EXTENT_NEED_WAIT
,
734 0, 0, &cached_state
, GFP_NOFS
);
735 err
= filemap_fdatawait_range(mapping
, start
, end
);
747 * when btree blocks are allocated, they have some corresponding bits set for
748 * them in one of two extent_io trees. This is used to make sure all of
749 * those extents are on disk for transaction or log commit
751 int btrfs_write_and_wait_marked_extents(struct btrfs_root
*root
,
752 struct extent_io_tree
*dirty_pages
, int mark
)
757 ret
= btrfs_write_marked_extents(root
, dirty_pages
, mark
);
758 ret2
= btrfs_wait_marked_extents(root
, dirty_pages
, mark
);
767 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
,
768 struct btrfs_root
*root
)
770 if (!trans
|| !trans
->transaction
) {
771 struct inode
*btree_inode
;
772 btree_inode
= root
->fs_info
->btree_inode
;
773 return filemap_write_and_wait(btree_inode
->i_mapping
);
775 return btrfs_write_and_wait_marked_extents(root
,
776 &trans
->transaction
->dirty_pages
,
781 * this is used to update the root pointer in the tree of tree roots.
783 * But, in the case of the extent allocation tree, updating the root
784 * pointer may allocate blocks which may change the root of the extent
787 * So, this loops and repeats and makes sure the cowonly root didn't
788 * change while the root pointer was being updated in the metadata.
790 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
791 struct btrfs_root
*root
)
796 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
798 old_root_used
= btrfs_root_used(&root
->root_item
);
799 btrfs_write_dirty_block_groups(trans
, root
);
802 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
803 if (old_root_bytenr
== root
->node
->start
&&
804 old_root_used
== btrfs_root_used(&root
->root_item
))
807 btrfs_set_root_node(&root
->root_item
, root
->node
);
808 ret
= btrfs_update_root(trans
, tree_root
,
814 old_root_used
= btrfs_root_used(&root
->root_item
);
815 ret
= btrfs_write_dirty_block_groups(trans
, root
);
820 if (root
!= root
->fs_info
->extent_root
)
821 switch_commit_root(root
);
827 * update all the cowonly tree roots on disk
829 * The error handling in this function may not be obvious. Any of the
830 * failures will cause the file system to go offline. We still need
831 * to clean up the delayed refs.
833 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
,
834 struct btrfs_root
*root
)
836 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
837 struct list_head
*next
;
838 struct extent_buffer
*eb
;
841 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
845 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
846 ret
= btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
,
848 btrfs_tree_unlock(eb
);
849 free_extent_buffer(eb
);
854 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
858 ret
= btrfs_run_dev_stats(trans
, root
->fs_info
);
860 ret
= btrfs_run_dev_replace(trans
, root
->fs_info
);
863 ret
= btrfs_run_qgroups(trans
, root
->fs_info
);
866 /* run_qgroups might have added some more refs */
867 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
870 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
871 next
= fs_info
->dirty_cowonly_roots
.next
;
873 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
875 ret
= update_cowonly_root(trans
, root
);
880 down_write(&fs_info
->extent_commit_sem
);
881 switch_commit_root(fs_info
->extent_root
);
882 up_write(&fs_info
->extent_commit_sem
);
884 btrfs_after_dev_replace_commit(fs_info
);
890 * dead roots are old snapshots that need to be deleted. This allocates
891 * a dirty root struct and adds it into the list of dead roots that need to
894 int btrfs_add_dead_root(struct btrfs_root
*root
)
896 spin_lock(&root
->fs_info
->trans_lock
);
897 list_add(&root
->root_list
, &root
->fs_info
->dead_roots
);
898 spin_unlock(&root
->fs_info
->trans_lock
);
903 * update all the cowonly tree roots on disk
905 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
,
906 struct btrfs_root
*root
)
908 struct btrfs_root
*gang
[8];
909 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
914 spin_lock(&fs_info
->fs_roots_radix_lock
);
916 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
919 BTRFS_ROOT_TRANS_TAG
);
922 for (i
= 0; i
< ret
; i
++) {
924 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
925 (unsigned long)root
->root_key
.objectid
,
926 BTRFS_ROOT_TRANS_TAG
);
927 spin_unlock(&fs_info
->fs_roots_radix_lock
);
929 btrfs_free_log(trans
, root
);
930 btrfs_update_reloc_root(trans
, root
);
931 btrfs_orphan_commit_root(trans
, root
);
933 btrfs_save_ino_cache(root
, trans
);
935 /* see comments in should_cow_block() */
939 if (root
->commit_root
!= root
->node
) {
940 mutex_lock(&root
->fs_commit_mutex
);
941 switch_commit_root(root
);
942 btrfs_unpin_free_ino(root
);
943 mutex_unlock(&root
->fs_commit_mutex
);
945 btrfs_set_root_node(&root
->root_item
,
949 err
= btrfs_update_root(trans
, fs_info
->tree_root
,
952 spin_lock(&fs_info
->fs_roots_radix_lock
);
957 spin_unlock(&fs_info
->fs_roots_radix_lock
);
962 * defrag a given btree.
963 * Every leaf in the btree is read and defragged.
965 int btrfs_defrag_root(struct btrfs_root
*root
)
967 struct btrfs_fs_info
*info
= root
->fs_info
;
968 struct btrfs_trans_handle
*trans
;
971 if (xchg(&root
->defrag_running
, 1))
975 trans
= btrfs_start_transaction(root
, 0);
977 return PTR_ERR(trans
);
979 ret
= btrfs_defrag_leaves(trans
, root
);
981 btrfs_end_transaction(trans
, root
);
982 btrfs_btree_balance_dirty(info
->tree_root
);
985 if (btrfs_fs_closing(root
->fs_info
) || ret
!= -EAGAIN
)
988 root
->defrag_running
= 0;
993 * new snapshots need to be created at a very specific time in the
994 * transaction commit. This does the actual creation
996 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
997 struct btrfs_fs_info
*fs_info
,
998 struct btrfs_pending_snapshot
*pending
)
1000 struct btrfs_key key
;
1001 struct btrfs_root_item
*new_root_item
;
1002 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1003 struct btrfs_root
*root
= pending
->root
;
1004 struct btrfs_root
*parent_root
;
1005 struct btrfs_block_rsv
*rsv
;
1006 struct inode
*parent_inode
;
1007 struct btrfs_path
*path
;
1008 struct btrfs_dir_item
*dir_item
;
1009 struct dentry
*parent
;
1010 struct dentry
*dentry
;
1011 struct extent_buffer
*tmp
;
1012 struct extent_buffer
*old
;
1013 struct timespec cur_time
= CURRENT_TIME
;
1021 path
= btrfs_alloc_path();
1023 ret
= pending
->error
= -ENOMEM
;
1024 goto path_alloc_fail
;
1027 new_root_item
= kmalloc(sizeof(*new_root_item
), GFP_NOFS
);
1028 if (!new_root_item
) {
1029 ret
= pending
->error
= -ENOMEM
;
1030 goto root_item_alloc_fail
;
1033 ret
= btrfs_find_free_objectid(tree_root
, &objectid
);
1035 pending
->error
= ret
;
1036 goto no_free_objectid
;
1039 btrfs_reloc_pre_snapshot(trans
, pending
, &to_reserve
);
1041 if (to_reserve
> 0) {
1042 ret
= btrfs_block_rsv_add(root
, &pending
->block_rsv
,
1044 BTRFS_RESERVE_NO_FLUSH
);
1046 pending
->error
= ret
;
1047 goto no_free_objectid
;
1051 ret
= btrfs_qgroup_inherit(trans
, fs_info
, root
->root_key
.objectid
,
1052 objectid
, pending
->inherit
);
1054 pending
->error
= ret
;
1055 goto no_free_objectid
;
1058 key
.objectid
= objectid
;
1059 key
.offset
= (u64
)-1;
1060 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1062 rsv
= trans
->block_rsv
;
1063 trans
->block_rsv
= &pending
->block_rsv
;
1065 dentry
= pending
->dentry
;
1066 parent
= dget_parent(dentry
);
1067 parent_inode
= parent
->d_inode
;
1068 parent_root
= BTRFS_I(parent_inode
)->root
;
1069 record_root_in_trans(trans
, parent_root
);
1072 * insert the directory item
1074 ret
= btrfs_set_inode_index(parent_inode
, &index
);
1075 BUG_ON(ret
); /* -ENOMEM */
1077 /* check if there is a file/dir which has the same name. */
1078 dir_item
= btrfs_lookup_dir_item(NULL
, parent_root
, path
,
1079 btrfs_ino(parent_inode
),
1080 dentry
->d_name
.name
,
1081 dentry
->d_name
.len
, 0);
1082 if (dir_item
!= NULL
&& !IS_ERR(dir_item
)) {
1083 pending
->error
= -EEXIST
;
1085 } else if (IS_ERR(dir_item
)) {
1086 ret
= PTR_ERR(dir_item
);
1087 btrfs_abort_transaction(trans
, root
, ret
);
1090 btrfs_release_path(path
);
1093 * pull in the delayed directory update
1094 * and the delayed inode item
1095 * otherwise we corrupt the FS during
1098 ret
= btrfs_run_delayed_items(trans
, root
);
1099 if (ret
) { /* Transaction aborted */
1100 btrfs_abort_transaction(trans
, root
, ret
);
1104 record_root_in_trans(trans
, root
);
1105 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
1106 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
1107 btrfs_check_and_init_root_item(new_root_item
);
1109 root_flags
= btrfs_root_flags(new_root_item
);
1110 if (pending
->readonly
)
1111 root_flags
|= BTRFS_ROOT_SUBVOL_RDONLY
;
1113 root_flags
&= ~BTRFS_ROOT_SUBVOL_RDONLY
;
1114 btrfs_set_root_flags(new_root_item
, root_flags
);
1116 btrfs_set_root_generation_v2(new_root_item
,
1118 uuid_le_gen(&new_uuid
);
1119 memcpy(new_root_item
->uuid
, new_uuid
.b
, BTRFS_UUID_SIZE
);
1120 memcpy(new_root_item
->parent_uuid
, root
->root_item
.uuid
,
1122 new_root_item
->otime
.sec
= cpu_to_le64(cur_time
.tv_sec
);
1123 new_root_item
->otime
.nsec
= cpu_to_le32(cur_time
.tv_nsec
);
1124 btrfs_set_root_otransid(new_root_item
, trans
->transid
);
1125 memset(&new_root_item
->stime
, 0, sizeof(new_root_item
->stime
));
1126 memset(&new_root_item
->rtime
, 0, sizeof(new_root_item
->rtime
));
1127 btrfs_set_root_stransid(new_root_item
, 0);
1128 btrfs_set_root_rtransid(new_root_item
, 0);
1130 old
= btrfs_lock_root_node(root
);
1131 ret
= btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
);
1133 btrfs_tree_unlock(old
);
1134 free_extent_buffer(old
);
1135 btrfs_abort_transaction(trans
, root
, ret
);
1139 btrfs_set_lock_blocking(old
);
1141 ret
= btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
1142 /* clean up in any case */
1143 btrfs_tree_unlock(old
);
1144 free_extent_buffer(old
);
1146 btrfs_abort_transaction(trans
, root
, ret
);
1150 /* see comments in should_cow_block() */
1151 root
->force_cow
= 1;
1154 btrfs_set_root_node(new_root_item
, tmp
);
1155 /* record when the snapshot was created in key.offset */
1156 key
.offset
= trans
->transid
;
1157 ret
= btrfs_insert_root(trans
, tree_root
, &key
, new_root_item
);
1158 btrfs_tree_unlock(tmp
);
1159 free_extent_buffer(tmp
);
1161 btrfs_abort_transaction(trans
, root
, ret
);
1166 * insert root back/forward references
1168 ret
= btrfs_add_root_ref(trans
, tree_root
, objectid
,
1169 parent_root
->root_key
.objectid
,
1170 btrfs_ino(parent_inode
), index
,
1171 dentry
->d_name
.name
, dentry
->d_name
.len
);
1173 btrfs_abort_transaction(trans
, root
, ret
);
1177 key
.offset
= (u64
)-1;
1178 pending
->snap
= btrfs_read_fs_root_no_name(root
->fs_info
, &key
);
1179 if (IS_ERR(pending
->snap
)) {
1180 ret
= PTR_ERR(pending
->snap
);
1181 btrfs_abort_transaction(trans
, root
, ret
);
1185 ret
= btrfs_reloc_post_snapshot(trans
, pending
);
1187 btrfs_abort_transaction(trans
, root
, ret
);
1191 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1193 btrfs_abort_transaction(trans
, root
, ret
);
1197 ret
= btrfs_insert_dir_item(trans
, parent_root
,
1198 dentry
->d_name
.name
, dentry
->d_name
.len
,
1200 BTRFS_FT_DIR
, index
);
1201 /* We have check then name at the beginning, so it is impossible. */
1202 BUG_ON(ret
== -EEXIST
|| ret
== -EOVERFLOW
);
1204 btrfs_abort_transaction(trans
, root
, ret
);
1208 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
1209 dentry
->d_name
.len
* 2);
1210 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
1211 ret
= btrfs_update_inode_fallback(trans
, parent_root
, parent_inode
);
1213 btrfs_abort_transaction(trans
, root
, ret
);
1216 trans
->block_rsv
= rsv
;
1218 kfree(new_root_item
);
1219 root_item_alloc_fail
:
1220 btrfs_free_path(path
);
1222 btrfs_block_rsv_release(root
, &pending
->block_rsv
, (u64
)-1);
1227 * create all the snapshots we've scheduled for creation
1229 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
,
1230 struct btrfs_fs_info
*fs_info
)
1232 struct btrfs_pending_snapshot
*pending
;
1233 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
1235 list_for_each_entry(pending
, head
, list
)
1236 create_pending_snapshot(trans
, fs_info
, pending
);
1240 static void update_super_roots(struct btrfs_root
*root
)
1242 struct btrfs_root_item
*root_item
;
1243 struct btrfs_super_block
*super
;
1245 super
= root
->fs_info
->super_copy
;
1247 root_item
= &root
->fs_info
->chunk_root
->root_item
;
1248 super
->chunk_root
= root_item
->bytenr
;
1249 super
->chunk_root_generation
= root_item
->generation
;
1250 super
->chunk_root_level
= root_item
->level
;
1252 root_item
= &root
->fs_info
->tree_root
->root_item
;
1253 super
->root
= root_item
->bytenr
;
1254 super
->generation
= root_item
->generation
;
1255 super
->root_level
= root_item
->level
;
1256 if (btrfs_test_opt(root
, SPACE_CACHE
))
1257 super
->cache_generation
= root_item
->generation
;
1260 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
1263 spin_lock(&info
->trans_lock
);
1264 if (info
->running_transaction
)
1265 ret
= info
->running_transaction
->in_commit
;
1266 spin_unlock(&info
->trans_lock
);
1270 int btrfs_transaction_blocked(struct btrfs_fs_info
*info
)
1273 spin_lock(&info
->trans_lock
);
1274 if (info
->running_transaction
)
1275 ret
= info
->running_transaction
->blocked
;
1276 spin_unlock(&info
->trans_lock
);
1281 * wait for the current transaction commit to start and block subsequent
1284 static void wait_current_trans_commit_start(struct btrfs_root
*root
,
1285 struct btrfs_transaction
*trans
)
1287 wait_event(root
->fs_info
->transaction_blocked_wait
, trans
->in_commit
);
1291 * wait for the current transaction to start and then become unblocked.
1294 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root
*root
,
1295 struct btrfs_transaction
*trans
)
1297 wait_event(root
->fs_info
->transaction_wait
,
1298 trans
->commit_done
|| (trans
->in_commit
&& !trans
->blocked
));
1302 * commit transactions asynchronously. once btrfs_commit_transaction_async
1303 * returns, any subsequent transaction will not be allowed to join.
1305 struct btrfs_async_commit
{
1306 struct btrfs_trans_handle
*newtrans
;
1307 struct btrfs_root
*root
;
1308 struct work_struct work
;
1311 static void do_async_commit(struct work_struct
*work
)
1313 struct btrfs_async_commit
*ac
=
1314 container_of(work
, struct btrfs_async_commit
, work
);
1317 * We've got freeze protection passed with the transaction.
1318 * Tell lockdep about it.
1320 if (ac
->newtrans
->type
< TRANS_JOIN_NOLOCK
)
1322 &ac
->root
->fs_info
->sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
1325 current
->journal_info
= ac
->newtrans
;
1327 btrfs_commit_transaction(ac
->newtrans
, ac
->root
);
1331 int btrfs_commit_transaction_async(struct btrfs_trans_handle
*trans
,
1332 struct btrfs_root
*root
,
1333 int wait_for_unblock
)
1335 struct btrfs_async_commit
*ac
;
1336 struct btrfs_transaction
*cur_trans
;
1338 ac
= kmalloc(sizeof(*ac
), GFP_NOFS
);
1342 INIT_WORK(&ac
->work
, do_async_commit
);
1344 ac
->newtrans
= btrfs_join_transaction(root
);
1345 if (IS_ERR(ac
->newtrans
)) {
1346 int err
= PTR_ERR(ac
->newtrans
);
1351 /* take transaction reference */
1352 cur_trans
= trans
->transaction
;
1353 atomic_inc(&cur_trans
->use_count
);
1355 btrfs_end_transaction(trans
, root
);
1358 * Tell lockdep we've released the freeze rwsem, since the
1359 * async commit thread will be the one to unlock it.
1361 if (trans
->type
< TRANS_JOIN_NOLOCK
)
1363 &root
->fs_info
->sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
1366 schedule_work(&ac
->work
);
1368 /* wait for transaction to start and unblock */
1369 if (wait_for_unblock
)
1370 wait_current_trans_commit_start_and_unblock(root
, cur_trans
);
1372 wait_current_trans_commit_start(root
, cur_trans
);
1374 if (current
->journal_info
== trans
)
1375 current
->journal_info
= NULL
;
1377 put_transaction(cur_trans
);
1382 static void cleanup_transaction(struct btrfs_trans_handle
*trans
,
1383 struct btrfs_root
*root
, int err
)
1385 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1387 WARN_ON(trans
->use_count
> 1);
1389 btrfs_abort_transaction(trans
, root
, err
);
1391 spin_lock(&root
->fs_info
->trans_lock
);
1392 list_del_init(&cur_trans
->list
);
1393 if (cur_trans
== root
->fs_info
->running_transaction
) {
1394 root
->fs_info
->running_transaction
= NULL
;
1395 root
->fs_info
->trans_no_join
= 0;
1397 spin_unlock(&root
->fs_info
->trans_lock
);
1399 btrfs_cleanup_one_transaction(trans
->transaction
, root
);
1401 put_transaction(cur_trans
);
1402 put_transaction(cur_trans
);
1404 trace_btrfs_transaction_commit(root
);
1406 btrfs_scrub_continue(root
);
1408 if (current
->journal_info
== trans
)
1409 current
->journal_info
= NULL
;
1411 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1414 static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle
*trans
,
1415 struct btrfs_root
*root
)
1417 int flush_on_commit
= btrfs_test_opt(root
, FLUSHONCOMMIT
);
1418 int snap_pending
= 0;
1421 if (!flush_on_commit
) {
1422 spin_lock(&root
->fs_info
->trans_lock
);
1423 if (!list_empty(&trans
->transaction
->pending_snapshots
))
1425 spin_unlock(&root
->fs_info
->trans_lock
);
1428 if (flush_on_commit
|| snap_pending
) {
1429 ret
= btrfs_start_delalloc_inodes(root
, 1);
1432 btrfs_wait_ordered_extents(root
, 1);
1435 ret
= btrfs_run_delayed_items(trans
, root
);
1440 * running the delayed items may have added new refs. account
1441 * them now so that they hinder processing of more delayed refs
1442 * as little as possible.
1444 btrfs_delayed_refs_qgroup_accounting(trans
, root
->fs_info
);
1447 * rename don't use btrfs_join_transaction, so, once we
1448 * set the transaction to blocked above, we aren't going
1449 * to get any new ordered operations. We can safely run
1450 * it here and no for sure that nothing new will be added
1453 ret
= btrfs_run_ordered_operations(root
, 1);
1459 * btrfs_transaction state sequence:
1460 * in_commit = 0, blocked = 0 (initial)
1461 * in_commit = 1, blocked = 1
1465 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
,
1466 struct btrfs_root
*root
)
1468 unsigned long joined
= 0;
1469 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1470 struct btrfs_transaction
*prev_trans
= NULL
;
1473 int should_grow
= 0;
1474 unsigned long now
= get_seconds();
1476 ret
= btrfs_run_ordered_operations(root
, 0);
1478 btrfs_abort_transaction(trans
, root
, ret
);
1479 btrfs_end_transaction(trans
, root
);
1483 /* Stop the commit early if ->aborted is set */
1484 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1485 ret
= cur_trans
->aborted
;
1486 btrfs_end_transaction(trans
, root
);
1490 /* make a pass through all the delayed refs we have so far
1491 * any runnings procs may add more while we are here
1493 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1495 btrfs_end_transaction(trans
, root
);
1499 btrfs_trans_release_metadata(trans
, root
);
1500 trans
->block_rsv
= NULL
;
1502 cur_trans
= trans
->transaction
;
1505 * set the flushing flag so procs in this transaction have to
1506 * start sending their work down.
1508 cur_trans
->delayed_refs
.flushing
= 1;
1510 if (!list_empty(&trans
->new_bgs
))
1511 btrfs_create_pending_block_groups(trans
, root
);
1513 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
1515 btrfs_end_transaction(trans
, root
);
1519 spin_lock(&cur_trans
->commit_lock
);
1520 if (cur_trans
->in_commit
) {
1521 spin_unlock(&cur_trans
->commit_lock
);
1522 atomic_inc(&cur_trans
->use_count
);
1523 ret
= btrfs_end_transaction(trans
, root
);
1525 wait_for_commit(root
, cur_trans
);
1527 put_transaction(cur_trans
);
1532 trans
->transaction
->in_commit
= 1;
1533 trans
->transaction
->blocked
= 1;
1534 spin_unlock(&cur_trans
->commit_lock
);
1535 wake_up(&root
->fs_info
->transaction_blocked_wait
);
1537 spin_lock(&root
->fs_info
->trans_lock
);
1538 if (cur_trans
->list
.prev
!= &root
->fs_info
->trans_list
) {
1539 prev_trans
= list_entry(cur_trans
->list
.prev
,
1540 struct btrfs_transaction
, list
);
1541 if (!prev_trans
->commit_done
) {
1542 atomic_inc(&prev_trans
->use_count
);
1543 spin_unlock(&root
->fs_info
->trans_lock
);
1545 wait_for_commit(root
, prev_trans
);
1547 put_transaction(prev_trans
);
1549 spin_unlock(&root
->fs_info
->trans_lock
);
1552 spin_unlock(&root
->fs_info
->trans_lock
);
1555 if (!btrfs_test_opt(root
, SSD
) &&
1556 (now
< cur_trans
->start_time
|| now
- cur_trans
->start_time
< 1))
1560 joined
= cur_trans
->num_joined
;
1562 WARN_ON(cur_trans
!= trans
->transaction
);
1564 ret
= btrfs_flush_all_pending_stuffs(trans
, root
);
1566 goto cleanup_transaction
;
1568 prepare_to_wait(&cur_trans
->writer_wait
, &wait
,
1569 TASK_UNINTERRUPTIBLE
);
1571 if (atomic_read(&cur_trans
->num_writers
) > 1)
1572 schedule_timeout(MAX_SCHEDULE_TIMEOUT
);
1573 else if (should_grow
)
1574 schedule_timeout(1);
1576 finish_wait(&cur_trans
->writer_wait
, &wait
);
1577 } while (atomic_read(&cur_trans
->num_writers
) > 1 ||
1578 (should_grow
&& cur_trans
->num_joined
!= joined
));
1580 ret
= btrfs_flush_all_pending_stuffs(trans
, root
);
1582 goto cleanup_transaction
;
1585 * Ok now we need to make sure to block out any other joins while we
1586 * commit the transaction. We could have started a join before setting
1587 * no_join so make sure to wait for num_writers to == 1 again.
1589 spin_lock(&root
->fs_info
->trans_lock
);
1590 root
->fs_info
->trans_no_join
= 1;
1591 spin_unlock(&root
->fs_info
->trans_lock
);
1592 wait_event(cur_trans
->writer_wait
,
1593 atomic_read(&cur_trans
->num_writers
) == 1);
1595 /* ->aborted might be set after the previous check, so check it */
1596 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1597 ret
= cur_trans
->aborted
;
1598 goto cleanup_transaction
;
1601 * the reloc mutex makes sure that we stop
1602 * the balancing code from coming in and moving
1603 * extents around in the middle of the commit
1605 mutex_lock(&root
->fs_info
->reloc_mutex
);
1608 * We needn't worry about the delayed items because we will
1609 * deal with them in create_pending_snapshot(), which is the
1610 * core function of the snapshot creation.
1612 ret
= create_pending_snapshots(trans
, root
->fs_info
);
1614 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1615 goto cleanup_transaction
;
1619 * We insert the dir indexes of the snapshots and update the inode
1620 * of the snapshots' parents after the snapshot creation, so there
1621 * are some delayed items which are not dealt with. Now deal with
1624 * We needn't worry that this operation will corrupt the snapshots,
1625 * because all the tree which are snapshoted will be forced to COW
1626 * the nodes and leaves.
1628 ret
= btrfs_run_delayed_items(trans
, root
);
1630 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1631 goto cleanup_transaction
;
1634 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long)-1);
1636 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1637 goto cleanup_transaction
;
1641 * make sure none of the code above managed to slip in a
1644 btrfs_assert_delayed_root_empty(root
);
1646 WARN_ON(cur_trans
!= trans
->transaction
);
1648 btrfs_scrub_pause(root
);
1649 /* btrfs_commit_tree_roots is responsible for getting the
1650 * various roots consistent with each other. Every pointer
1651 * in the tree of tree roots has to point to the most up to date
1652 * root for every subvolume and other tree. So, we have to keep
1653 * the tree logging code from jumping in and changing any
1656 * At this point in the commit, there can't be any tree-log
1657 * writers, but a little lower down we drop the trans mutex
1658 * and let new people in. By holding the tree_log_mutex
1659 * from now until after the super is written, we avoid races
1660 * with the tree-log code.
1662 mutex_lock(&root
->fs_info
->tree_log_mutex
);
1664 ret
= commit_fs_roots(trans
, root
);
1666 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1667 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1668 goto cleanup_transaction
;
1671 /* commit_fs_roots gets rid of all the tree log roots, it is now
1672 * safe to free the root of tree log roots
1674 btrfs_free_log_root_tree(trans
, root
->fs_info
);
1676 ret
= commit_cowonly_roots(trans
, root
);
1678 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1679 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1680 goto cleanup_transaction
;
1684 * The tasks which save the space cache and inode cache may also
1685 * update ->aborted, check it.
1687 if (unlikely(ACCESS_ONCE(cur_trans
->aborted
))) {
1688 ret
= cur_trans
->aborted
;
1689 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1690 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1691 goto cleanup_transaction
;
1694 btrfs_prepare_extent_commit(trans
, root
);
1696 cur_trans
= root
->fs_info
->running_transaction
;
1698 btrfs_set_root_node(&root
->fs_info
->tree_root
->root_item
,
1699 root
->fs_info
->tree_root
->node
);
1700 switch_commit_root(root
->fs_info
->tree_root
);
1702 btrfs_set_root_node(&root
->fs_info
->chunk_root
->root_item
,
1703 root
->fs_info
->chunk_root
->node
);
1704 switch_commit_root(root
->fs_info
->chunk_root
);
1706 assert_qgroups_uptodate(trans
);
1707 update_super_roots(root
);
1709 if (!root
->fs_info
->log_root_recovering
) {
1710 btrfs_set_super_log_root(root
->fs_info
->super_copy
, 0);
1711 btrfs_set_super_log_root_level(root
->fs_info
->super_copy
, 0);
1714 memcpy(root
->fs_info
->super_for_commit
, root
->fs_info
->super_copy
,
1715 sizeof(*root
->fs_info
->super_copy
));
1717 trans
->transaction
->blocked
= 0;
1718 spin_lock(&root
->fs_info
->trans_lock
);
1719 root
->fs_info
->running_transaction
= NULL
;
1720 root
->fs_info
->trans_no_join
= 0;
1721 spin_unlock(&root
->fs_info
->trans_lock
);
1722 mutex_unlock(&root
->fs_info
->reloc_mutex
);
1724 wake_up(&root
->fs_info
->transaction_wait
);
1726 ret
= btrfs_write_and_wait_transaction(trans
, root
);
1728 btrfs_error(root
->fs_info
, ret
,
1729 "Error while writing out transaction.");
1730 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1731 goto cleanup_transaction
;
1734 ret
= write_ctree_super(trans
, root
, 0);
1736 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1737 goto cleanup_transaction
;
1741 * the super is written, we can safely allow the tree-loggers
1742 * to go about their business
1744 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
1746 btrfs_finish_extent_commit(trans
, root
);
1748 cur_trans
->commit_done
= 1;
1750 root
->fs_info
->last_trans_committed
= cur_trans
->transid
;
1752 wake_up(&cur_trans
->commit_wait
);
1754 spin_lock(&root
->fs_info
->trans_lock
);
1755 list_del_init(&cur_trans
->list
);
1756 spin_unlock(&root
->fs_info
->trans_lock
);
1758 put_transaction(cur_trans
);
1759 put_transaction(cur_trans
);
1761 if (trans
->type
< TRANS_JOIN_NOLOCK
)
1762 sb_end_intwrite(root
->fs_info
->sb
);
1764 trace_btrfs_transaction_commit(root
);
1766 btrfs_scrub_continue(root
);
1768 if (current
->journal_info
== trans
)
1769 current
->journal_info
= NULL
;
1771 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1773 if (current
!= root
->fs_info
->transaction_kthread
)
1774 btrfs_run_delayed_iputs(root
);
1778 cleanup_transaction
:
1779 btrfs_trans_release_metadata(trans
, root
);
1780 trans
->block_rsv
= NULL
;
1781 btrfs_printk(root
->fs_info
, "Skipping commit of aborted transaction.\n");
1783 if (current
->journal_info
== trans
)
1784 current
->journal_info
= NULL
;
1785 cleanup_transaction(trans
, root
, ret
);
1791 * interface function to delete all the snapshots we have scheduled for deletion
1793 int btrfs_clean_old_snapshots(struct btrfs_root
*root
)
1796 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1798 spin_lock(&fs_info
->trans_lock
);
1799 list_splice_init(&fs_info
->dead_roots
, &list
);
1800 spin_unlock(&fs_info
->trans_lock
);
1802 while (!list_empty(&list
)) {
1805 root
= list_entry(list
.next
, struct btrfs_root
, root_list
);
1806 list_del(&root
->root_list
);
1808 btrfs_kill_all_delayed_nodes(root
);
1810 if (btrfs_header_backref_rev(root
->node
) <
1811 BTRFS_MIXED_BACKREF_REV
)
1812 ret
= btrfs_drop_snapshot(root
, NULL
, 0, 0);
1814 ret
=btrfs_drop_snapshot(root
, NULL
, 1, 0);