1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
7 #include <linux/slab.h>
8 #include <linux/sched.h>
9 #include <linux/sched/mm.h>
10 #include <linux/writeback.h>
11 #include <linux/pagemap.h>
12 #include <linux/blkdev.h>
13 #include <linux/uuid.h>
14 #include <linux/timekeeping.h>
18 #include "transaction.h"
22 #include "dev-replace.h"
24 #include "block-group.h"
25 #include "space-info.h"
28 #include "accessors.h"
29 #include "extent-tree.h"
30 #include "root-tree.h"
33 #include "uuid-tree.h"
35 #include "relocation.h"
38 static struct kmem_cache
*btrfs_trans_handle_cachep
;
40 #define BTRFS_ROOT_TRANS_TAG 0
43 * Transaction states and transitions
45 * No running transaction (fs tree blocks are not modified)
48 * | Call start_transaction() variants. Except btrfs_join_transaction_nostart().
50 * Transaction N [[TRANS_STATE_RUNNING]]
52 * | New trans handles can be attached to transaction N by calling all
53 * | start_transaction() variants.
56 * | Call btrfs_commit_transaction() on any trans handle attached to
59 * Transaction N [[TRANS_STATE_COMMIT_PREP]]
61 * | If there are simultaneous calls to btrfs_commit_transaction() one will win
62 * | the race and the rest will wait for the winner to commit the transaction.
64 * | The winner will wait for previous running transaction to completely finish
67 * Transaction N [[TRANS_STATE_COMMIT_START]]
69 * | Then one of the following happens:
70 * | - Wait for all other trans handle holders to release.
71 * | The btrfs_commit_transaction() caller will do the commit work.
72 * | - Wait for current transaction to be committed by others.
73 * | Other btrfs_commit_transaction() caller will do the commit work.
75 * | At this stage, only btrfs_join_transaction*() variants can attach
76 * | to this running transaction.
77 * | All other variants will wait for current one to finish and attach to
81 * | Caller is chosen to commit transaction N, and all other trans handle
82 * | haven been released.
84 * Transaction N [[TRANS_STATE_COMMIT_DOING]]
86 * | The heavy lifting transaction work is started.
87 * | From running delayed refs (modifying extent tree) to creating pending
88 * | snapshots, running qgroups.
89 * | In short, modify supporting trees to reflect modifications of subvolume
92 * | At this stage, all start_transaction() calls will wait for this
93 * | transaction to finish and attach to transaction N+1.
96 * | Until all supporting trees are updated.
98 * Transaction N [[TRANS_STATE_UNBLOCKED]]
100 * | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]]
101 * | need to write them back to disk and update |
104 * | At this stage, new transaction is allowed to |
106 * | All new start_transaction() calls will be |
107 * | attached to transid N+1. |
110 * | Until all tree blocks are super blocks are |
111 * | written to block devices |
113 * Transaction N [[TRANS_STATE_COMPLETED]] V
114 * All tree blocks and super blocks are written. Transaction N+1
115 * This transaction is finished and all its [[TRANS_STATE_COMMIT_START]]
116 * data structures will be cleaned up. | Life goes on
118 static const unsigned int btrfs_blocked_trans_types
[TRANS_STATE_MAX
] = {
119 [TRANS_STATE_RUNNING
] = 0U,
120 [TRANS_STATE_COMMIT_PREP
] = 0U,
121 [TRANS_STATE_COMMIT_START
] = (__TRANS_START
| __TRANS_ATTACH
),
122 [TRANS_STATE_COMMIT_DOING
] = (__TRANS_START
|
125 __TRANS_JOIN_NOSTART
),
126 [TRANS_STATE_UNBLOCKED
] = (__TRANS_START
|
129 __TRANS_JOIN_NOLOCK
|
130 __TRANS_JOIN_NOSTART
),
131 [TRANS_STATE_SUPER_COMMITTED
] = (__TRANS_START
|
134 __TRANS_JOIN_NOLOCK
|
135 __TRANS_JOIN_NOSTART
),
136 [TRANS_STATE_COMPLETED
] = (__TRANS_START
|
139 __TRANS_JOIN_NOLOCK
|
140 __TRANS_JOIN_NOSTART
),
143 void btrfs_put_transaction(struct btrfs_transaction
*transaction
)
145 WARN_ON(refcount_read(&transaction
->use_count
) == 0);
146 if (refcount_dec_and_test(&transaction
->use_count
)) {
147 BUG_ON(!list_empty(&transaction
->list
));
148 WARN_ON(!RB_EMPTY_ROOT(
149 &transaction
->delayed_refs
.href_root
.rb_root
));
150 WARN_ON(!RB_EMPTY_ROOT(
151 &transaction
->delayed_refs
.dirty_extent_root
));
152 if (transaction
->delayed_refs
.pending_csums
)
153 btrfs_err(transaction
->fs_info
,
154 "pending csums is %llu",
155 transaction
->delayed_refs
.pending_csums
);
157 * If any block groups are found in ->deleted_bgs then it's
158 * because the transaction was aborted and a commit did not
159 * happen (things failed before writing the new superblock
160 * and calling btrfs_finish_extent_commit()), so we can not
161 * discard the physical locations of the block groups.
163 while (!list_empty(&transaction
->deleted_bgs
)) {
164 struct btrfs_block_group
*cache
;
166 cache
= list_first_entry(&transaction
->deleted_bgs
,
167 struct btrfs_block_group
,
169 list_del_init(&cache
->bg_list
);
170 btrfs_unfreeze_block_group(cache
);
171 btrfs_put_block_group(cache
);
173 WARN_ON(!list_empty(&transaction
->dev_update_list
));
178 static noinline
void switch_commit_roots(struct btrfs_trans_handle
*trans
)
180 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
181 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
182 struct btrfs_root
*root
, *tmp
;
185 * At this point no one can be using this transaction to modify any tree
186 * and no one can start another transaction to modify any tree either.
188 ASSERT(cur_trans
->state
== TRANS_STATE_COMMIT_DOING
);
190 down_write(&fs_info
->commit_root_sem
);
192 if (test_bit(BTRFS_FS_RELOC_RUNNING
, &fs_info
->flags
))
193 fs_info
->last_reloc_trans
= trans
->transid
;
195 list_for_each_entry_safe(root
, tmp
, &cur_trans
->switch_commits
,
197 list_del_init(&root
->dirty_list
);
198 free_extent_buffer(root
->commit_root
);
199 root
->commit_root
= btrfs_root_node(root
);
200 extent_io_tree_release(&root
->dirty_log_pages
);
201 btrfs_qgroup_clean_swapped_blocks(root
);
204 /* We can free old roots now. */
205 spin_lock(&cur_trans
->dropped_roots_lock
);
206 while (!list_empty(&cur_trans
->dropped_roots
)) {
207 root
= list_first_entry(&cur_trans
->dropped_roots
,
208 struct btrfs_root
, root_list
);
209 list_del_init(&root
->root_list
);
210 spin_unlock(&cur_trans
->dropped_roots_lock
);
211 btrfs_free_log(trans
, root
);
212 btrfs_drop_and_free_fs_root(fs_info
, root
);
213 spin_lock(&cur_trans
->dropped_roots_lock
);
215 spin_unlock(&cur_trans
->dropped_roots_lock
);
217 up_write(&fs_info
->commit_root_sem
);
220 static inline void extwriter_counter_inc(struct btrfs_transaction
*trans
,
223 if (type
& TRANS_EXTWRITERS
)
224 atomic_inc(&trans
->num_extwriters
);
227 static inline void extwriter_counter_dec(struct btrfs_transaction
*trans
,
230 if (type
& TRANS_EXTWRITERS
)
231 atomic_dec(&trans
->num_extwriters
);
234 static inline void extwriter_counter_init(struct btrfs_transaction
*trans
,
237 atomic_set(&trans
->num_extwriters
, ((type
& TRANS_EXTWRITERS
) ? 1 : 0));
240 static inline int extwriter_counter_read(struct btrfs_transaction
*trans
)
242 return atomic_read(&trans
->num_extwriters
);
246 * To be called after doing the chunk btree updates right after allocating a new
247 * chunk (after btrfs_chunk_alloc_add_chunk_item() is called), when removing a
248 * chunk after all chunk btree updates and after finishing the second phase of
249 * chunk allocation (btrfs_create_pending_block_groups()) in case some block
250 * group had its chunk item insertion delayed to the second phase.
252 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle
*trans
)
254 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
256 if (!trans
->chunk_bytes_reserved
)
259 btrfs_block_rsv_release(fs_info
, &fs_info
->chunk_block_rsv
,
260 trans
->chunk_bytes_reserved
, NULL
);
261 trans
->chunk_bytes_reserved
= 0;
265 * either allocate a new transaction or hop into the existing one
267 static noinline
int join_transaction(struct btrfs_fs_info
*fs_info
,
270 struct btrfs_transaction
*cur_trans
;
272 spin_lock(&fs_info
->trans_lock
);
274 /* The file system has been taken offline. No new transactions. */
275 if (BTRFS_FS_ERROR(fs_info
)) {
276 spin_unlock(&fs_info
->trans_lock
);
280 cur_trans
= fs_info
->running_transaction
;
282 if (TRANS_ABORTED(cur_trans
)) {
283 spin_unlock(&fs_info
->trans_lock
);
284 return cur_trans
->aborted
;
286 if (btrfs_blocked_trans_types
[cur_trans
->state
] & type
) {
287 spin_unlock(&fs_info
->trans_lock
);
290 refcount_inc(&cur_trans
->use_count
);
291 atomic_inc(&cur_trans
->num_writers
);
292 extwriter_counter_inc(cur_trans
, type
);
293 spin_unlock(&fs_info
->trans_lock
);
294 btrfs_lockdep_acquire(fs_info
, btrfs_trans_num_writers
);
295 btrfs_lockdep_acquire(fs_info
, btrfs_trans_num_extwriters
);
298 spin_unlock(&fs_info
->trans_lock
);
301 * If we are ATTACH or TRANS_JOIN_NOSTART, we just want to catch the
302 * current transaction, and commit it. If there is no transaction, just
305 if (type
== TRANS_ATTACH
|| type
== TRANS_JOIN_NOSTART
)
309 * JOIN_NOLOCK only happens during the transaction commit, so
310 * it is impossible that ->running_transaction is NULL
312 BUG_ON(type
== TRANS_JOIN_NOLOCK
);
314 cur_trans
= kmalloc(sizeof(*cur_trans
), GFP_NOFS
);
318 btrfs_lockdep_acquire(fs_info
, btrfs_trans_num_writers
);
319 btrfs_lockdep_acquire(fs_info
, btrfs_trans_num_extwriters
);
321 spin_lock(&fs_info
->trans_lock
);
322 if (fs_info
->running_transaction
) {
324 * someone started a transaction after we unlocked. Make sure
325 * to redo the checks above
327 btrfs_lockdep_release(fs_info
, btrfs_trans_num_extwriters
);
328 btrfs_lockdep_release(fs_info
, btrfs_trans_num_writers
);
331 } else if (BTRFS_FS_ERROR(fs_info
)) {
332 spin_unlock(&fs_info
->trans_lock
);
333 btrfs_lockdep_release(fs_info
, btrfs_trans_num_extwriters
);
334 btrfs_lockdep_release(fs_info
, btrfs_trans_num_writers
);
339 cur_trans
->fs_info
= fs_info
;
340 atomic_set(&cur_trans
->pending_ordered
, 0);
341 init_waitqueue_head(&cur_trans
->pending_wait
);
342 atomic_set(&cur_trans
->num_writers
, 1);
343 extwriter_counter_init(cur_trans
, type
);
344 init_waitqueue_head(&cur_trans
->writer_wait
);
345 init_waitqueue_head(&cur_trans
->commit_wait
);
346 cur_trans
->state
= TRANS_STATE_RUNNING
;
348 * One for this trans handle, one so it will live on until we
349 * commit the transaction.
351 refcount_set(&cur_trans
->use_count
, 2);
352 cur_trans
->flags
= 0;
353 cur_trans
->start_time
= ktime_get_seconds();
355 memset(&cur_trans
->delayed_refs
, 0, sizeof(cur_trans
->delayed_refs
));
357 cur_trans
->delayed_refs
.href_root
= RB_ROOT_CACHED
;
358 cur_trans
->delayed_refs
.dirty_extent_root
= RB_ROOT
;
359 atomic_set(&cur_trans
->delayed_refs
.num_entries
, 0);
362 * although the tree mod log is per file system and not per transaction,
363 * the log must never go across transaction boundaries.
366 if (!list_empty(&fs_info
->tree_mod_seq_list
))
367 WARN(1, KERN_ERR
"BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n");
368 if (!RB_EMPTY_ROOT(&fs_info
->tree_mod_log
))
369 WARN(1, KERN_ERR
"BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n");
370 atomic64_set(&fs_info
->tree_mod_seq
, 0);
372 spin_lock_init(&cur_trans
->delayed_refs
.lock
);
374 INIT_LIST_HEAD(&cur_trans
->pending_snapshots
);
375 INIT_LIST_HEAD(&cur_trans
->dev_update_list
);
376 INIT_LIST_HEAD(&cur_trans
->switch_commits
);
377 INIT_LIST_HEAD(&cur_trans
->dirty_bgs
);
378 INIT_LIST_HEAD(&cur_trans
->io_bgs
);
379 INIT_LIST_HEAD(&cur_trans
->dropped_roots
);
380 mutex_init(&cur_trans
->cache_write_mutex
);
381 spin_lock_init(&cur_trans
->dirty_bgs_lock
);
382 INIT_LIST_HEAD(&cur_trans
->deleted_bgs
);
383 spin_lock_init(&cur_trans
->dropped_roots_lock
);
384 list_add_tail(&cur_trans
->list
, &fs_info
->trans_list
);
385 extent_io_tree_init(fs_info
, &cur_trans
->dirty_pages
,
386 IO_TREE_TRANS_DIRTY_PAGES
);
387 extent_io_tree_init(fs_info
, &cur_trans
->pinned_extents
,
388 IO_TREE_FS_PINNED_EXTENTS
);
389 fs_info
->generation
++;
390 cur_trans
->transid
= fs_info
->generation
;
391 fs_info
->running_transaction
= cur_trans
;
392 cur_trans
->aborted
= 0;
393 spin_unlock(&fs_info
->trans_lock
);
399 * This does all the record keeping required to make sure that a shareable root
400 * is properly recorded in a given transaction. This is required to make sure
401 * the old root from before we joined the transaction is deleted when the
402 * transaction commits.
404 static int record_root_in_trans(struct btrfs_trans_handle
*trans
,
405 struct btrfs_root
*root
,
408 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
411 if ((test_bit(BTRFS_ROOT_SHAREABLE
, &root
->state
) &&
412 root
->last_trans
< trans
->transid
) || force
) {
413 WARN_ON(!force
&& root
->commit_root
!= root
->node
);
416 * see below for IN_TRANS_SETUP usage rules
417 * we have the reloc mutex held now, so there
418 * is only one writer in this function
420 set_bit(BTRFS_ROOT_IN_TRANS_SETUP
, &root
->state
);
422 /* make sure readers find IN_TRANS_SETUP before
423 * they find our root->last_trans update
427 spin_lock(&fs_info
->fs_roots_radix_lock
);
428 if (root
->last_trans
== trans
->transid
&& !force
) {
429 spin_unlock(&fs_info
->fs_roots_radix_lock
);
432 radix_tree_tag_set(&fs_info
->fs_roots_radix
,
433 (unsigned long)root
->root_key
.objectid
,
434 BTRFS_ROOT_TRANS_TAG
);
435 spin_unlock(&fs_info
->fs_roots_radix_lock
);
436 root
->last_trans
= trans
->transid
;
438 /* this is pretty tricky. We don't want to
439 * take the relocation lock in btrfs_record_root_in_trans
440 * unless we're really doing the first setup for this root in
443 * Normally we'd use root->last_trans as a flag to decide
444 * if we want to take the expensive mutex.
446 * But, we have to set root->last_trans before we
447 * init the relocation root, otherwise, we trip over warnings
448 * in ctree.c. The solution used here is to flag ourselves
449 * with root IN_TRANS_SETUP. When this is 1, we're still
450 * fixing up the reloc trees and everyone must wait.
452 * When this is zero, they can trust root->last_trans and fly
453 * through btrfs_record_root_in_trans without having to take the
454 * lock. smp_wmb() makes sure that all the writes above are
455 * done before we pop in the zero below
457 ret
= btrfs_init_reloc_root(trans
, root
);
458 smp_mb__before_atomic();
459 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP
, &root
->state
);
465 void btrfs_add_dropped_root(struct btrfs_trans_handle
*trans
,
466 struct btrfs_root
*root
)
468 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
469 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
471 /* Add ourselves to the transaction dropped list */
472 spin_lock(&cur_trans
->dropped_roots_lock
);
473 list_add_tail(&root
->root_list
, &cur_trans
->dropped_roots
);
474 spin_unlock(&cur_trans
->dropped_roots_lock
);
476 /* Make sure we don't try to update the root at commit time */
477 spin_lock(&fs_info
->fs_roots_radix_lock
);
478 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
479 (unsigned long)root
->root_key
.objectid
,
480 BTRFS_ROOT_TRANS_TAG
);
481 spin_unlock(&fs_info
->fs_roots_radix_lock
);
484 int btrfs_record_root_in_trans(struct btrfs_trans_handle
*trans
,
485 struct btrfs_root
*root
)
487 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
490 if (!test_bit(BTRFS_ROOT_SHAREABLE
, &root
->state
))
494 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
498 if (root
->last_trans
== trans
->transid
&&
499 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP
, &root
->state
))
502 mutex_lock(&fs_info
->reloc_mutex
);
503 ret
= record_root_in_trans(trans
, root
, 0);
504 mutex_unlock(&fs_info
->reloc_mutex
);
509 static inline int is_transaction_blocked(struct btrfs_transaction
*trans
)
511 return (trans
->state
>= TRANS_STATE_COMMIT_START
&&
512 trans
->state
< TRANS_STATE_UNBLOCKED
&&
513 !TRANS_ABORTED(trans
));
516 /* wait for commit against the current transaction to become unblocked
517 * when this is done, it is safe to start a new transaction, but the current
518 * transaction might not be fully on disk.
520 static void wait_current_trans(struct btrfs_fs_info
*fs_info
)
522 struct btrfs_transaction
*cur_trans
;
524 spin_lock(&fs_info
->trans_lock
);
525 cur_trans
= fs_info
->running_transaction
;
526 if (cur_trans
&& is_transaction_blocked(cur_trans
)) {
527 refcount_inc(&cur_trans
->use_count
);
528 spin_unlock(&fs_info
->trans_lock
);
530 btrfs_might_wait_for_state(fs_info
, BTRFS_LOCKDEP_TRANS_UNBLOCKED
);
531 wait_event(fs_info
->transaction_wait
,
532 cur_trans
->state
>= TRANS_STATE_UNBLOCKED
||
533 TRANS_ABORTED(cur_trans
));
534 btrfs_put_transaction(cur_trans
);
536 spin_unlock(&fs_info
->trans_lock
);
540 static int may_wait_transaction(struct btrfs_fs_info
*fs_info
, int type
)
542 if (test_bit(BTRFS_FS_LOG_RECOVERING
, &fs_info
->flags
))
545 if (type
== TRANS_START
)
551 static inline bool need_reserve_reloc_root(struct btrfs_root
*root
)
553 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
555 if (!fs_info
->reloc_ctl
||
556 !test_bit(BTRFS_ROOT_SHAREABLE
, &root
->state
) ||
557 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
564 static struct btrfs_trans_handle
*
565 start_transaction(struct btrfs_root
*root
, unsigned int num_items
,
566 unsigned int type
, enum btrfs_reserve_flush_enum flush
,
567 bool enforce_qgroups
)
569 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
570 struct btrfs_block_rsv
*delayed_refs_rsv
= &fs_info
->delayed_refs_rsv
;
571 struct btrfs_trans_handle
*h
;
572 struct btrfs_transaction
*cur_trans
;
574 u64 qgroup_reserved
= 0;
575 bool reloc_reserved
= false;
576 bool do_chunk_alloc
= false;
579 if (BTRFS_FS_ERROR(fs_info
))
580 return ERR_PTR(-EROFS
);
582 if (current
->journal_info
) {
583 WARN_ON(type
& TRANS_EXTWRITERS
);
584 h
= current
->journal_info
;
585 refcount_inc(&h
->use_count
);
586 WARN_ON(refcount_read(&h
->use_count
) > 2);
587 h
->orig_rsv
= h
->block_rsv
;
593 * Do the reservation before we join the transaction so we can do all
594 * the appropriate flushing if need be.
596 if (num_items
&& root
!= fs_info
->chunk_root
) {
597 struct btrfs_block_rsv
*rsv
= &fs_info
->trans_block_rsv
;
598 u64 delayed_refs_bytes
= 0;
600 qgroup_reserved
= num_items
* fs_info
->nodesize
;
602 * Use prealloc for now, as there might be a currently running
603 * transaction that could free this reserved space prematurely
606 ret
= btrfs_qgroup_reserve_meta_prealloc(root
, qgroup_reserved
,
607 enforce_qgroups
, false);
612 * We want to reserve all the bytes we may need all at once, so
613 * we only do 1 enospc flushing cycle per transaction start. We
614 * accomplish this by simply assuming we'll do num_items worth
615 * of delayed refs updates in this trans handle, and refill that
616 * amount for whatever is missing in the reserve.
618 num_bytes
= btrfs_calc_insert_metadata_size(fs_info
, num_items
);
619 if (flush
== BTRFS_RESERVE_FLUSH_ALL
&&
620 !btrfs_block_rsv_full(delayed_refs_rsv
)) {
621 delayed_refs_bytes
= btrfs_calc_delayed_ref_bytes(fs_info
,
623 num_bytes
+= delayed_refs_bytes
;
627 * Do the reservation for the relocation root creation
629 if (need_reserve_reloc_root(root
)) {
630 num_bytes
+= fs_info
->nodesize
;
631 reloc_reserved
= true;
634 ret
= btrfs_reserve_metadata_bytes(fs_info
, rsv
, num_bytes
, flush
);
637 if (delayed_refs_bytes
) {
638 btrfs_migrate_to_delayed_refs_rsv(fs_info
, delayed_refs_bytes
);
639 num_bytes
-= delayed_refs_bytes
;
641 btrfs_block_rsv_add_bytes(rsv
, num_bytes
, true);
643 if (rsv
->space_info
->force_alloc
)
644 do_chunk_alloc
= true;
645 } else if (num_items
== 0 && flush
== BTRFS_RESERVE_FLUSH_ALL
&&
646 !btrfs_block_rsv_full(delayed_refs_rsv
)) {
648 * Some people call with btrfs_start_transaction(root, 0)
649 * because they can be throttled, but have some other mechanism
650 * for reserving space. We still want these guys to refill the
651 * delayed block_rsv so just add 1 items worth of reservation
654 ret
= btrfs_delayed_refs_rsv_refill(fs_info
, flush
);
659 h
= kmem_cache_zalloc(btrfs_trans_handle_cachep
, GFP_NOFS
);
666 * If we are JOIN_NOLOCK we're already committing a transaction and
667 * waiting on this guy, so we don't need to do the sb_start_intwrite
668 * because we're already holding a ref. We need this because we could
669 * have raced in and did an fsync() on a file which can kick a commit
670 * and then we deadlock with somebody doing a freeze.
672 * If we are ATTACH, it means we just want to catch the current
673 * transaction and commit it, so we needn't do sb_start_intwrite().
675 if (type
& __TRANS_FREEZABLE
)
676 sb_start_intwrite(fs_info
->sb
);
678 if (may_wait_transaction(fs_info
, type
))
679 wait_current_trans(fs_info
);
682 ret
= join_transaction(fs_info
, type
);
684 wait_current_trans(fs_info
);
685 if (unlikely(type
== TRANS_ATTACH
||
686 type
== TRANS_JOIN_NOSTART
))
689 } while (ret
== -EBUSY
);
694 cur_trans
= fs_info
->running_transaction
;
696 h
->transid
= cur_trans
->transid
;
697 h
->transaction
= cur_trans
;
698 refcount_set(&h
->use_count
, 1);
699 h
->fs_info
= root
->fs_info
;
702 INIT_LIST_HEAD(&h
->new_bgs
);
705 if (cur_trans
->state
>= TRANS_STATE_COMMIT_START
&&
706 may_wait_transaction(fs_info
, type
)) {
707 current
->journal_info
= h
;
708 btrfs_commit_transaction(h
);
713 trace_btrfs_space_reservation(fs_info
, "transaction",
714 h
->transid
, num_bytes
, 1);
715 h
->block_rsv
= &fs_info
->trans_block_rsv
;
716 h
->bytes_reserved
= num_bytes
;
717 h
->reloc_reserved
= reloc_reserved
;
721 * Now that we have found a transaction to be a part of, convert the
722 * qgroup reservation from prealloc to pertrans. A different transaction
723 * can't race in and free our pertrans out from under us.
726 btrfs_qgroup_convert_reserved_meta(root
, qgroup_reserved
);
729 if (!current
->journal_info
)
730 current
->journal_info
= h
;
733 * If the space_info is marked ALLOC_FORCE then we'll get upgraded to
734 * ALLOC_FORCE the first run through, and then we won't allocate for
735 * anybody else who races in later. We don't care about the return
738 if (do_chunk_alloc
&& num_bytes
) {
739 u64 flags
= h
->block_rsv
->space_info
->flags
;
741 btrfs_chunk_alloc(h
, btrfs_get_alloc_profile(fs_info
, flags
),
742 CHUNK_ALLOC_NO_FORCE
);
746 * btrfs_record_root_in_trans() needs to alloc new extents, and may
747 * call btrfs_join_transaction() while we're also starting a
750 * Thus it need to be called after current->journal_info initialized,
751 * or we can deadlock.
753 ret
= btrfs_record_root_in_trans(h
, root
);
756 * The transaction handle is fully initialized and linked with
757 * other structures so it needs to be ended in case of errors,
760 btrfs_end_transaction(h
);
767 if (type
& __TRANS_FREEZABLE
)
768 sb_end_intwrite(fs_info
->sb
);
769 kmem_cache_free(btrfs_trans_handle_cachep
, h
);
772 btrfs_block_rsv_release(fs_info
, &fs_info
->trans_block_rsv
,
775 btrfs_qgroup_free_meta_prealloc(root
, qgroup_reserved
);
779 struct btrfs_trans_handle
*btrfs_start_transaction(struct btrfs_root
*root
,
780 unsigned int num_items
)
782 return start_transaction(root
, num_items
, TRANS_START
,
783 BTRFS_RESERVE_FLUSH_ALL
, true);
786 struct btrfs_trans_handle
*btrfs_start_transaction_fallback_global_rsv(
787 struct btrfs_root
*root
,
788 unsigned int num_items
)
790 return start_transaction(root
, num_items
, TRANS_START
,
791 BTRFS_RESERVE_FLUSH_ALL_STEAL
, false);
794 struct btrfs_trans_handle
*btrfs_join_transaction(struct btrfs_root
*root
)
796 return start_transaction(root
, 0, TRANS_JOIN
, BTRFS_RESERVE_NO_FLUSH
,
800 struct btrfs_trans_handle
*btrfs_join_transaction_spacecache(struct btrfs_root
*root
)
802 return start_transaction(root
, 0, TRANS_JOIN_NOLOCK
,
803 BTRFS_RESERVE_NO_FLUSH
, true);
807 * Similar to regular join but it never starts a transaction when none is
808 * running or when there's a running one at a state >= TRANS_STATE_UNBLOCKED.
809 * This is similar to btrfs_attach_transaction() but it allows the join to
810 * happen if the transaction commit already started but it's not yet in the
811 * "doing" phase (the state is < TRANS_STATE_COMMIT_DOING).
813 struct btrfs_trans_handle
*btrfs_join_transaction_nostart(struct btrfs_root
*root
)
815 return start_transaction(root
, 0, TRANS_JOIN_NOSTART
,
816 BTRFS_RESERVE_NO_FLUSH
, true);
820 * btrfs_attach_transaction() - catch the running transaction
822 * It is used when we want to commit the current the transaction, but
823 * don't want to start a new one.
825 * Note: If this function return -ENOENT, it just means there is no
826 * running transaction. But it is possible that the inactive transaction
827 * is still in the memory, not fully on disk. If you hope there is no
828 * inactive transaction in the fs when -ENOENT is returned, you should
830 * btrfs_attach_transaction_barrier()
832 struct btrfs_trans_handle
*btrfs_attach_transaction(struct btrfs_root
*root
)
834 return start_transaction(root
, 0, TRANS_ATTACH
,
835 BTRFS_RESERVE_NO_FLUSH
, true);
839 * btrfs_attach_transaction_barrier() - catch the running transaction
841 * It is similar to the above function, the difference is this one
842 * will wait for all the inactive transactions until they fully
845 struct btrfs_trans_handle
*
846 btrfs_attach_transaction_barrier(struct btrfs_root
*root
)
848 struct btrfs_trans_handle
*trans
;
850 trans
= start_transaction(root
, 0, TRANS_ATTACH
,
851 BTRFS_RESERVE_NO_FLUSH
, true);
852 if (trans
== ERR_PTR(-ENOENT
)) {
855 ret
= btrfs_wait_for_commit(root
->fs_info
, 0);
863 /* Wait for a transaction commit to reach at least the given state. */
864 static noinline
void wait_for_commit(struct btrfs_transaction
*commit
,
865 const enum btrfs_trans_state min_state
)
867 struct btrfs_fs_info
*fs_info
= commit
->fs_info
;
868 u64 transid
= commit
->transid
;
872 * At the moment this function is called with min_state either being
873 * TRANS_STATE_COMPLETED or TRANS_STATE_SUPER_COMMITTED.
875 if (min_state
== TRANS_STATE_COMPLETED
)
876 btrfs_might_wait_for_state(fs_info
, BTRFS_LOCKDEP_TRANS_COMPLETED
);
878 btrfs_might_wait_for_state(fs_info
, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED
);
881 wait_event(commit
->commit_wait
, commit
->state
>= min_state
);
883 btrfs_put_transaction(commit
);
885 if (min_state
< TRANS_STATE_COMPLETED
)
889 * A transaction isn't really completed until all of the
890 * previous transactions are completed, but with fsync we can
891 * end up with SUPER_COMMITTED transactions before a COMPLETED
892 * transaction. Wait for those.
895 spin_lock(&fs_info
->trans_lock
);
896 commit
= list_first_entry_or_null(&fs_info
->trans_list
,
897 struct btrfs_transaction
,
899 if (!commit
|| commit
->transid
> transid
) {
900 spin_unlock(&fs_info
->trans_lock
);
903 refcount_inc(&commit
->use_count
);
905 spin_unlock(&fs_info
->trans_lock
);
909 int btrfs_wait_for_commit(struct btrfs_fs_info
*fs_info
, u64 transid
)
911 struct btrfs_transaction
*cur_trans
= NULL
, *t
;
915 if (transid
<= fs_info
->last_trans_committed
)
918 /* find specified transaction */
919 spin_lock(&fs_info
->trans_lock
);
920 list_for_each_entry(t
, &fs_info
->trans_list
, list
) {
921 if (t
->transid
== transid
) {
923 refcount_inc(&cur_trans
->use_count
);
927 if (t
->transid
> transid
) {
932 spin_unlock(&fs_info
->trans_lock
);
935 * The specified transaction doesn't exist, or we
936 * raced with btrfs_commit_transaction
939 if (transid
> fs_info
->last_trans_committed
)
944 /* find newest transaction that is committing | committed */
945 spin_lock(&fs_info
->trans_lock
);
946 list_for_each_entry_reverse(t
, &fs_info
->trans_list
,
948 if (t
->state
>= TRANS_STATE_COMMIT_START
) {
949 if (t
->state
== TRANS_STATE_COMPLETED
)
952 refcount_inc(&cur_trans
->use_count
);
956 spin_unlock(&fs_info
->trans_lock
);
958 goto out
; /* nothing committing|committed */
961 wait_for_commit(cur_trans
, TRANS_STATE_COMPLETED
);
962 ret
= cur_trans
->aborted
;
963 btrfs_put_transaction(cur_trans
);
968 void btrfs_throttle(struct btrfs_fs_info
*fs_info
)
970 wait_current_trans(fs_info
);
973 bool btrfs_should_end_transaction(struct btrfs_trans_handle
*trans
)
975 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
977 if (cur_trans
->state
>= TRANS_STATE_COMMIT_START
||
978 test_bit(BTRFS_DELAYED_REFS_FLUSHING
, &cur_trans
->delayed_refs
.flags
))
981 if (btrfs_check_space_for_delayed_refs(trans
->fs_info
))
984 return !!btrfs_block_rsv_check(&trans
->fs_info
->global_block_rsv
, 50);
987 static void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
)
990 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
992 if (!trans
->block_rsv
) {
993 ASSERT(!trans
->bytes_reserved
);
997 if (!trans
->bytes_reserved
)
1000 ASSERT(trans
->block_rsv
== &fs_info
->trans_block_rsv
);
1001 trace_btrfs_space_reservation(fs_info
, "transaction",
1002 trans
->transid
, trans
->bytes_reserved
, 0);
1003 btrfs_block_rsv_release(fs_info
, trans
->block_rsv
,
1004 trans
->bytes_reserved
, NULL
);
1005 trans
->bytes_reserved
= 0;
1008 static int __btrfs_end_transaction(struct btrfs_trans_handle
*trans
,
1011 struct btrfs_fs_info
*info
= trans
->fs_info
;
1012 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
1015 if (refcount_read(&trans
->use_count
) > 1) {
1016 refcount_dec(&trans
->use_count
);
1017 trans
->block_rsv
= trans
->orig_rsv
;
1021 btrfs_trans_release_metadata(trans
);
1022 trans
->block_rsv
= NULL
;
1024 btrfs_create_pending_block_groups(trans
);
1026 btrfs_trans_release_chunk_metadata(trans
);
1028 if (trans
->type
& __TRANS_FREEZABLE
)
1029 sb_end_intwrite(info
->sb
);
1031 WARN_ON(cur_trans
!= info
->running_transaction
);
1032 WARN_ON(atomic_read(&cur_trans
->num_writers
) < 1);
1033 atomic_dec(&cur_trans
->num_writers
);
1034 extwriter_counter_dec(cur_trans
, trans
->type
);
1036 cond_wake_up(&cur_trans
->writer_wait
);
1038 btrfs_lockdep_release(info
, btrfs_trans_num_extwriters
);
1039 btrfs_lockdep_release(info
, btrfs_trans_num_writers
);
1041 btrfs_put_transaction(cur_trans
);
1043 if (current
->journal_info
== trans
)
1044 current
->journal_info
= NULL
;
1047 btrfs_run_delayed_iputs(info
);
1049 if (TRANS_ABORTED(trans
) || BTRFS_FS_ERROR(info
)) {
1050 wake_up_process(info
->transaction_kthread
);
1051 if (TRANS_ABORTED(trans
))
1052 err
= trans
->aborted
;
1057 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
1061 int btrfs_end_transaction(struct btrfs_trans_handle
*trans
)
1063 return __btrfs_end_transaction(trans
, 0);
1066 int btrfs_end_transaction_throttle(struct btrfs_trans_handle
*trans
)
1068 return __btrfs_end_transaction(trans
, 1);
1072 * when btree blocks are allocated, they have some corresponding bits set for
1073 * them in one of two extent_io trees. This is used to make sure all of
1074 * those extents are sent to disk but does not wait on them
1076 int btrfs_write_marked_extents(struct btrfs_fs_info
*fs_info
,
1077 struct extent_io_tree
*dirty_pages
, int mark
)
1081 struct address_space
*mapping
= fs_info
->btree_inode
->i_mapping
;
1082 struct extent_state
*cached_state
= NULL
;
1086 while (find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
1087 mark
, &cached_state
)) {
1088 bool wait_writeback
= false;
1090 err
= convert_extent_bit(dirty_pages
, start
, end
,
1092 mark
, &cached_state
);
1094 * convert_extent_bit can return -ENOMEM, which is most of the
1095 * time a temporary error. So when it happens, ignore the error
1096 * and wait for writeback of this range to finish - because we
1097 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
1098 * to __btrfs_wait_marked_extents() would not know that
1099 * writeback for this range started and therefore wouldn't
1100 * wait for it to finish - we don't want to commit a
1101 * superblock that points to btree nodes/leafs for which
1102 * writeback hasn't finished yet (and without errors).
1103 * We cleanup any entries left in the io tree when committing
1104 * the transaction (through extent_io_tree_release()).
1106 if (err
== -ENOMEM
) {
1108 wait_writeback
= true;
1111 err
= filemap_fdatawrite_range(mapping
, start
, end
);
1114 else if (wait_writeback
)
1115 werr
= filemap_fdatawait_range(mapping
, start
, end
);
1116 free_extent_state(cached_state
);
1117 cached_state
= NULL
;
1125 * when btree blocks are allocated, they have some corresponding bits set for
1126 * them in one of two extent_io trees. This is used to make sure all of
1127 * those extents are on disk for transaction or log commit. We wait
1128 * on all the pages and clear them from the dirty pages state tree
1130 static int __btrfs_wait_marked_extents(struct btrfs_fs_info
*fs_info
,
1131 struct extent_io_tree
*dirty_pages
)
1135 struct address_space
*mapping
= fs_info
->btree_inode
->i_mapping
;
1136 struct extent_state
*cached_state
= NULL
;
1140 while (find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
1141 EXTENT_NEED_WAIT
, &cached_state
)) {
1143 * Ignore -ENOMEM errors returned by clear_extent_bit().
1144 * When committing the transaction, we'll remove any entries
1145 * left in the io tree. For a log commit, we don't remove them
1146 * after committing the log because the tree can be accessed
1147 * concurrently - we do it only at transaction commit time when
1148 * it's safe to do it (through extent_io_tree_release()).
1150 err
= clear_extent_bit(dirty_pages
, start
, end
,
1151 EXTENT_NEED_WAIT
, &cached_state
);
1155 err
= filemap_fdatawait_range(mapping
, start
, end
);
1158 free_extent_state(cached_state
);
1159 cached_state
= NULL
;
1168 static int btrfs_wait_extents(struct btrfs_fs_info
*fs_info
,
1169 struct extent_io_tree
*dirty_pages
)
1171 bool errors
= false;
1174 err
= __btrfs_wait_marked_extents(fs_info
, dirty_pages
);
1175 if (test_and_clear_bit(BTRFS_FS_BTREE_ERR
, &fs_info
->flags
))
1183 int btrfs_wait_tree_log_extents(struct btrfs_root
*log_root
, int mark
)
1185 struct btrfs_fs_info
*fs_info
= log_root
->fs_info
;
1186 struct extent_io_tree
*dirty_pages
= &log_root
->dirty_log_pages
;
1187 bool errors
= false;
1190 ASSERT(log_root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
);
1192 err
= __btrfs_wait_marked_extents(fs_info
, dirty_pages
);
1193 if ((mark
& EXTENT_DIRTY
) &&
1194 test_and_clear_bit(BTRFS_FS_LOG1_ERR
, &fs_info
->flags
))
1197 if ((mark
& EXTENT_NEW
) &&
1198 test_and_clear_bit(BTRFS_FS_LOG2_ERR
, &fs_info
->flags
))
1207 * When btree blocks are allocated the corresponding extents are marked dirty.
1208 * This function ensures such extents are persisted on disk for transaction or
1211 * @trans: transaction whose dirty pages we'd like to write
1213 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle
*trans
)
1217 struct extent_io_tree
*dirty_pages
= &trans
->transaction
->dirty_pages
;
1218 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1219 struct blk_plug plug
;
1221 blk_start_plug(&plug
);
1222 ret
= btrfs_write_marked_extents(fs_info
, dirty_pages
, EXTENT_DIRTY
);
1223 blk_finish_plug(&plug
);
1224 ret2
= btrfs_wait_extents(fs_info
, dirty_pages
);
1226 extent_io_tree_release(&trans
->transaction
->dirty_pages
);
1237 * this is used to update the root pointer in the tree of tree roots.
1239 * But, in the case of the extent allocation tree, updating the root
1240 * pointer may allocate blocks which may change the root of the extent
1243 * So, this loops and repeats and makes sure the cowonly root didn't
1244 * change while the root pointer was being updated in the metadata.
1246 static int update_cowonly_root(struct btrfs_trans_handle
*trans
,
1247 struct btrfs_root
*root
)
1250 u64 old_root_bytenr
;
1252 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1253 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1255 old_root_used
= btrfs_root_used(&root
->root_item
);
1258 old_root_bytenr
= btrfs_root_bytenr(&root
->root_item
);
1259 if (old_root_bytenr
== root
->node
->start
&&
1260 old_root_used
== btrfs_root_used(&root
->root_item
))
1263 btrfs_set_root_node(&root
->root_item
, root
->node
);
1264 ret
= btrfs_update_root(trans
, tree_root
,
1270 old_root_used
= btrfs_root_used(&root
->root_item
);
1277 * update all the cowonly tree roots on disk
1279 * The error handling in this function may not be obvious. Any of the
1280 * failures will cause the file system to go offline. We still need
1281 * to clean up the delayed refs.
1283 static noinline
int commit_cowonly_roots(struct btrfs_trans_handle
*trans
)
1285 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1286 struct list_head
*dirty_bgs
= &trans
->transaction
->dirty_bgs
;
1287 struct list_head
*io_bgs
= &trans
->transaction
->io_bgs
;
1288 struct list_head
*next
;
1289 struct extent_buffer
*eb
;
1293 * At this point no one can be using this transaction to modify any tree
1294 * and no one can start another transaction to modify any tree either.
1296 ASSERT(trans
->transaction
->state
== TRANS_STATE_COMMIT_DOING
);
1298 eb
= btrfs_lock_root_node(fs_info
->tree_root
);
1299 ret
= btrfs_cow_block(trans
, fs_info
->tree_root
, eb
, NULL
,
1300 0, &eb
, BTRFS_NESTING_COW
);
1301 btrfs_tree_unlock(eb
);
1302 free_extent_buffer(eb
);
1307 ret
= btrfs_run_dev_stats(trans
);
1310 ret
= btrfs_run_dev_replace(trans
);
1313 ret
= btrfs_run_qgroups(trans
);
1317 ret
= btrfs_setup_space_cache(trans
);
1322 while (!list_empty(&fs_info
->dirty_cowonly_roots
)) {
1323 struct btrfs_root
*root
;
1324 next
= fs_info
->dirty_cowonly_roots
.next
;
1325 list_del_init(next
);
1326 root
= list_entry(next
, struct btrfs_root
, dirty_list
);
1327 clear_bit(BTRFS_ROOT_DIRTY
, &root
->state
);
1329 list_add_tail(&root
->dirty_list
,
1330 &trans
->transaction
->switch_commits
);
1331 ret
= update_cowonly_root(trans
, root
);
1336 /* Now flush any delayed refs generated by updating all of the roots */
1337 ret
= btrfs_run_delayed_refs(trans
, (unsigned long)-1);
1341 while (!list_empty(dirty_bgs
) || !list_empty(io_bgs
)) {
1342 ret
= btrfs_write_dirty_block_groups(trans
);
1347 * We're writing the dirty block groups, which could generate
1348 * delayed refs, which could generate more dirty block groups,
1349 * so we want to keep this flushing in this loop to make sure
1350 * everything gets run.
1352 ret
= btrfs_run_delayed_refs(trans
, (unsigned long)-1);
1357 if (!list_empty(&fs_info
->dirty_cowonly_roots
))
1360 /* Update dev-replace pointer once everything is committed */
1361 fs_info
->dev_replace
.committed_cursor_left
=
1362 fs_info
->dev_replace
.cursor_left_last_write_of_item
;
1368 * If we had a pending drop we need to see if there are any others left in our
1369 * dead roots list, and if not clear our bit and wake any waiters.
1371 void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info
*fs_info
)
1374 * We put the drop in progress roots at the front of the list, so if the
1375 * first entry doesn't have UNFINISHED_DROP set we can wake everybody
1378 spin_lock(&fs_info
->trans_lock
);
1379 if (!list_empty(&fs_info
->dead_roots
)) {
1380 struct btrfs_root
*root
= list_first_entry(&fs_info
->dead_roots
,
1383 if (test_bit(BTRFS_ROOT_UNFINISHED_DROP
, &root
->state
)) {
1384 spin_unlock(&fs_info
->trans_lock
);
1388 spin_unlock(&fs_info
->trans_lock
);
1390 btrfs_wake_unfinished_drop(fs_info
);
1394 * dead roots are old snapshots that need to be deleted. This allocates
1395 * a dirty root struct and adds it into the list of dead roots that need to
1398 void btrfs_add_dead_root(struct btrfs_root
*root
)
1400 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1402 spin_lock(&fs_info
->trans_lock
);
1403 if (list_empty(&root
->root_list
)) {
1404 btrfs_grab_root(root
);
1406 /* We want to process the partially complete drops first. */
1407 if (test_bit(BTRFS_ROOT_UNFINISHED_DROP
, &root
->state
))
1408 list_add(&root
->root_list
, &fs_info
->dead_roots
);
1410 list_add_tail(&root
->root_list
, &fs_info
->dead_roots
);
1412 spin_unlock(&fs_info
->trans_lock
);
1416 * Update each subvolume root and its relocation root, if it exists, in the tree
1417 * of tree roots. Also free log roots if they exist.
1419 static noinline
int commit_fs_roots(struct btrfs_trans_handle
*trans
)
1421 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1422 struct btrfs_root
*gang
[8];
1427 * At this point no one can be using this transaction to modify any tree
1428 * and no one can start another transaction to modify any tree either.
1430 ASSERT(trans
->transaction
->state
== TRANS_STATE_COMMIT_DOING
);
1432 spin_lock(&fs_info
->fs_roots_radix_lock
);
1434 ret
= radix_tree_gang_lookup_tag(&fs_info
->fs_roots_radix
,
1437 BTRFS_ROOT_TRANS_TAG
);
1440 for (i
= 0; i
< ret
; i
++) {
1441 struct btrfs_root
*root
= gang
[i
];
1445 * At this point we can neither have tasks logging inodes
1446 * from a root nor trying to commit a log tree.
1448 ASSERT(atomic_read(&root
->log_writers
) == 0);
1449 ASSERT(atomic_read(&root
->log_commit
[0]) == 0);
1450 ASSERT(atomic_read(&root
->log_commit
[1]) == 0);
1452 radix_tree_tag_clear(&fs_info
->fs_roots_radix
,
1453 (unsigned long)root
->root_key
.objectid
,
1454 BTRFS_ROOT_TRANS_TAG
);
1455 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1457 btrfs_free_log(trans
, root
);
1458 ret2
= btrfs_update_reloc_root(trans
, root
);
1462 /* see comments in should_cow_block() */
1463 clear_bit(BTRFS_ROOT_FORCE_COW
, &root
->state
);
1464 smp_mb__after_atomic();
1466 if (root
->commit_root
!= root
->node
) {
1467 list_add_tail(&root
->dirty_list
,
1468 &trans
->transaction
->switch_commits
);
1469 btrfs_set_root_node(&root
->root_item
,
1473 ret2
= btrfs_update_root(trans
, fs_info
->tree_root
,
1478 spin_lock(&fs_info
->fs_roots_radix_lock
);
1479 btrfs_qgroup_free_meta_all_pertrans(root
);
1482 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1487 * defrag a given btree.
1488 * Every leaf in the btree is read and defragged.
1490 int btrfs_defrag_root(struct btrfs_root
*root
)
1492 struct btrfs_fs_info
*info
= root
->fs_info
;
1493 struct btrfs_trans_handle
*trans
;
1496 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING
, &root
->state
))
1500 trans
= btrfs_start_transaction(root
, 0);
1501 if (IS_ERR(trans
)) {
1502 ret
= PTR_ERR(trans
);
1506 ret
= btrfs_defrag_leaves(trans
, root
);
1508 btrfs_end_transaction(trans
);
1509 btrfs_btree_balance_dirty(info
);
1512 if (btrfs_fs_closing(info
) || ret
!= -EAGAIN
)
1515 if (btrfs_defrag_cancelled(info
)) {
1516 btrfs_debug(info
, "defrag_root cancelled");
1521 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING
, &root
->state
);
1526 * Do all special snapshot related qgroup dirty hack.
1528 * Will do all needed qgroup inherit and dirty hack like switch commit
1529 * roots inside one transaction and write all btree into disk, to make
1532 static int qgroup_account_snapshot(struct btrfs_trans_handle
*trans
,
1533 struct btrfs_root
*src
,
1534 struct btrfs_root
*parent
,
1535 struct btrfs_qgroup_inherit
*inherit
,
1538 struct btrfs_fs_info
*fs_info
= src
->fs_info
;
1542 * Save some performance in the case that qgroups are not
1543 * enabled. If this check races with the ioctl, rescan will
1546 if (!test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
))
1550 * Ensure dirty @src will be committed. Or, after coming
1551 * commit_fs_roots() and switch_commit_roots(), any dirty but not
1552 * recorded root will never be updated again, causing an outdated root
1555 ret
= record_root_in_trans(trans
, src
, 1);
1560 * btrfs_qgroup_inherit relies on a consistent view of the usage for the
1561 * src root, so we must run the delayed refs here.
1563 * However this isn't particularly fool proof, because there's no
1564 * synchronization keeping us from changing the tree after this point
1565 * before we do the qgroup_inherit, or even from making changes while
1566 * we're doing the qgroup_inherit. But that's a problem for the future,
1567 * for now flush the delayed refs to narrow the race window where the
1568 * qgroup counters could end up wrong.
1570 ret
= btrfs_run_delayed_refs(trans
, (unsigned long)-1);
1572 btrfs_abort_transaction(trans
, ret
);
1576 ret
= commit_fs_roots(trans
);
1579 ret
= btrfs_qgroup_account_extents(trans
);
1583 /* Now qgroup are all updated, we can inherit it to new qgroups */
1584 ret
= btrfs_qgroup_inherit(trans
, src
->root_key
.objectid
, dst_objectid
,
1590 * Now we do a simplified commit transaction, which will:
1591 * 1) commit all subvolume and extent tree
1592 * To ensure all subvolume and extent tree have a valid
1593 * commit_root to accounting later insert_dir_item()
1594 * 2) write all btree blocks onto disk
1595 * This is to make sure later btree modification will be cowed
1596 * Or commit_root can be populated and cause wrong qgroup numbers
1597 * In this simplified commit, we don't really care about other trees
1598 * like chunk and root tree, as they won't affect qgroup.
1599 * And we don't write super to avoid half committed status.
1601 ret
= commit_cowonly_roots(trans
);
1604 switch_commit_roots(trans
);
1605 ret
= btrfs_write_and_wait_transaction(trans
);
1607 btrfs_handle_fs_error(fs_info
, ret
,
1608 "Error while writing out transaction for qgroup");
1612 * Force parent root to be updated, as we recorded it before so its
1613 * last_trans == cur_transid.
1614 * Or it won't be committed again onto disk after later
1618 ret
= record_root_in_trans(trans
, parent
, 1);
1623 * new snapshots need to be created at a very specific time in the
1624 * transaction commit. This does the actual creation.
1627 * If the error which may affect the commitment of the current transaction
1628 * happens, we should return the error number. If the error which just affect
1629 * the creation of the pending snapshots, just return 0.
1631 static noinline
int create_pending_snapshot(struct btrfs_trans_handle
*trans
,
1632 struct btrfs_pending_snapshot
*pending
)
1635 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1636 struct btrfs_key key
;
1637 struct btrfs_root_item
*new_root_item
;
1638 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1639 struct btrfs_root
*root
= pending
->root
;
1640 struct btrfs_root
*parent_root
;
1641 struct btrfs_block_rsv
*rsv
;
1642 struct inode
*parent_inode
= pending
->dir
;
1643 struct btrfs_path
*path
;
1644 struct btrfs_dir_item
*dir_item
;
1645 struct extent_buffer
*tmp
;
1646 struct extent_buffer
*old
;
1647 struct timespec64 cur_time
;
1653 unsigned int nofs_flags
;
1654 struct fscrypt_name fname
;
1656 ASSERT(pending
->path
);
1657 path
= pending
->path
;
1659 ASSERT(pending
->root_item
);
1660 new_root_item
= pending
->root_item
;
1663 * We're inside a transaction and must make sure that any potential
1664 * allocations with GFP_KERNEL in fscrypt won't recurse back to
1667 nofs_flags
= memalloc_nofs_save();
1668 pending
->error
= fscrypt_setup_filename(parent_inode
,
1669 &pending
->dentry
->d_name
, 0,
1671 memalloc_nofs_restore(nofs_flags
);
1675 pending
->error
= btrfs_get_free_objectid(tree_root
, &objectid
);
1680 * Make qgroup to skip current new snapshot's qgroupid, as it is
1681 * accounted by later btrfs_qgroup_inherit().
1683 btrfs_set_skip_qgroup(trans
, objectid
);
1685 btrfs_reloc_pre_snapshot(pending
, &to_reserve
);
1687 if (to_reserve
> 0) {
1688 pending
->error
= btrfs_block_rsv_add(fs_info
,
1689 &pending
->block_rsv
,
1691 BTRFS_RESERVE_NO_FLUSH
);
1693 goto clear_skip_qgroup
;
1696 key
.objectid
= objectid
;
1697 key
.offset
= (u64
)-1;
1698 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1700 rsv
= trans
->block_rsv
;
1701 trans
->block_rsv
= &pending
->block_rsv
;
1702 trans
->bytes_reserved
= trans
->block_rsv
->reserved
;
1703 trace_btrfs_space_reservation(fs_info
, "transaction",
1705 trans
->bytes_reserved
, 1);
1706 parent_root
= BTRFS_I(parent_inode
)->root
;
1707 ret
= record_root_in_trans(trans
, parent_root
, 0);
1710 cur_time
= current_time(parent_inode
);
1713 * insert the directory item
1715 ret
= btrfs_set_inode_index(BTRFS_I(parent_inode
), &index
);
1717 btrfs_abort_transaction(trans
, ret
);
1721 /* check if there is a file/dir which has the same name. */
1722 dir_item
= btrfs_lookup_dir_item(NULL
, parent_root
, path
,
1723 btrfs_ino(BTRFS_I(parent_inode
)),
1724 &fname
.disk_name
, 0);
1725 if (dir_item
!= NULL
&& !IS_ERR(dir_item
)) {
1726 pending
->error
= -EEXIST
;
1727 goto dir_item_existed
;
1728 } else if (IS_ERR(dir_item
)) {
1729 ret
= PTR_ERR(dir_item
);
1730 btrfs_abort_transaction(trans
, ret
);
1733 btrfs_release_path(path
);
1736 * pull in the delayed directory update
1737 * and the delayed inode item
1738 * otherwise we corrupt the FS during
1741 ret
= btrfs_run_delayed_items(trans
);
1742 if (ret
) { /* Transaction aborted */
1743 btrfs_abort_transaction(trans
, ret
);
1747 ret
= record_root_in_trans(trans
, root
, 0);
1749 btrfs_abort_transaction(trans
, ret
);
1752 btrfs_set_root_last_snapshot(&root
->root_item
, trans
->transid
);
1753 memcpy(new_root_item
, &root
->root_item
, sizeof(*new_root_item
));
1754 btrfs_check_and_init_root_item(new_root_item
);
1756 root_flags
= btrfs_root_flags(new_root_item
);
1757 if (pending
->readonly
)
1758 root_flags
|= BTRFS_ROOT_SUBVOL_RDONLY
;
1760 root_flags
&= ~BTRFS_ROOT_SUBVOL_RDONLY
;
1761 btrfs_set_root_flags(new_root_item
, root_flags
);
1763 btrfs_set_root_generation_v2(new_root_item
,
1765 generate_random_guid(new_root_item
->uuid
);
1766 memcpy(new_root_item
->parent_uuid
, root
->root_item
.uuid
,
1768 if (!(root_flags
& BTRFS_ROOT_SUBVOL_RDONLY
)) {
1769 memset(new_root_item
->received_uuid
, 0,
1770 sizeof(new_root_item
->received_uuid
));
1771 memset(&new_root_item
->stime
, 0, sizeof(new_root_item
->stime
));
1772 memset(&new_root_item
->rtime
, 0, sizeof(new_root_item
->rtime
));
1773 btrfs_set_root_stransid(new_root_item
, 0);
1774 btrfs_set_root_rtransid(new_root_item
, 0);
1776 btrfs_set_stack_timespec_sec(&new_root_item
->otime
, cur_time
.tv_sec
);
1777 btrfs_set_stack_timespec_nsec(&new_root_item
->otime
, cur_time
.tv_nsec
);
1778 btrfs_set_root_otransid(new_root_item
, trans
->transid
);
1780 old
= btrfs_lock_root_node(root
);
1781 ret
= btrfs_cow_block(trans
, root
, old
, NULL
, 0, &old
,
1784 btrfs_tree_unlock(old
);
1785 free_extent_buffer(old
);
1786 btrfs_abort_transaction(trans
, ret
);
1790 ret
= btrfs_copy_root(trans
, root
, old
, &tmp
, objectid
);
1791 /* clean up in any case */
1792 btrfs_tree_unlock(old
);
1793 free_extent_buffer(old
);
1795 btrfs_abort_transaction(trans
, ret
);
1798 /* see comments in should_cow_block() */
1799 set_bit(BTRFS_ROOT_FORCE_COW
, &root
->state
);
1802 btrfs_set_root_node(new_root_item
, tmp
);
1803 /* record when the snapshot was created in key.offset */
1804 key
.offset
= trans
->transid
;
1805 ret
= btrfs_insert_root(trans
, tree_root
, &key
, new_root_item
);
1806 btrfs_tree_unlock(tmp
);
1807 free_extent_buffer(tmp
);
1809 btrfs_abort_transaction(trans
, ret
);
1814 * insert root back/forward references
1816 ret
= btrfs_add_root_ref(trans
, objectid
,
1817 parent_root
->root_key
.objectid
,
1818 btrfs_ino(BTRFS_I(parent_inode
)), index
,
1821 btrfs_abort_transaction(trans
, ret
);
1825 key
.offset
= (u64
)-1;
1826 pending
->snap
= btrfs_get_new_fs_root(fs_info
, objectid
, pending
->anon_dev
);
1827 if (IS_ERR(pending
->snap
)) {
1828 ret
= PTR_ERR(pending
->snap
);
1829 pending
->snap
= NULL
;
1830 btrfs_abort_transaction(trans
, ret
);
1834 ret
= btrfs_reloc_post_snapshot(trans
, pending
);
1836 btrfs_abort_transaction(trans
, ret
);
1841 * Do special qgroup accounting for snapshot, as we do some qgroup
1842 * snapshot hack to do fast snapshot.
1843 * To co-operate with that hack, we do hack again.
1844 * Or snapshot will be greatly slowed down by a subtree qgroup rescan
1846 ret
= qgroup_account_snapshot(trans
, root
, parent_root
,
1847 pending
->inherit
, objectid
);
1851 ret
= btrfs_insert_dir_item(trans
, &fname
.disk_name
,
1852 BTRFS_I(parent_inode
), &key
, BTRFS_FT_DIR
,
1854 /* We have check then name at the beginning, so it is impossible. */
1855 BUG_ON(ret
== -EEXIST
|| ret
== -EOVERFLOW
);
1857 btrfs_abort_transaction(trans
, ret
);
1861 btrfs_i_size_write(BTRFS_I(parent_inode
), parent_inode
->i_size
+
1862 fname
.disk_name
.len
* 2);
1863 parent_inode
->i_mtime
= inode_set_ctime_current(parent_inode
);
1864 ret
= btrfs_update_inode_fallback(trans
, parent_root
, BTRFS_I(parent_inode
));
1866 btrfs_abort_transaction(trans
, ret
);
1869 ret
= btrfs_uuid_tree_add(trans
, new_root_item
->uuid
,
1870 BTRFS_UUID_KEY_SUBVOL
,
1873 btrfs_abort_transaction(trans
, ret
);
1876 if (!btrfs_is_empty_uuid(new_root_item
->received_uuid
)) {
1877 ret
= btrfs_uuid_tree_add(trans
, new_root_item
->received_uuid
,
1878 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
1880 if (ret
&& ret
!= -EEXIST
) {
1881 btrfs_abort_transaction(trans
, ret
);
1887 pending
->error
= ret
;
1889 trans
->block_rsv
= rsv
;
1890 trans
->bytes_reserved
= 0;
1892 btrfs_clear_skip_qgroup(trans
);
1894 fscrypt_free_filename(&fname
);
1896 kfree(new_root_item
);
1897 pending
->root_item
= NULL
;
1898 btrfs_free_path(path
);
1899 pending
->path
= NULL
;
1905 * create all the snapshots we've scheduled for creation
1907 static noinline
int create_pending_snapshots(struct btrfs_trans_handle
*trans
)
1909 struct btrfs_pending_snapshot
*pending
, *next
;
1910 struct list_head
*head
= &trans
->transaction
->pending_snapshots
;
1913 list_for_each_entry_safe(pending
, next
, head
, list
) {
1914 list_del(&pending
->list
);
1915 ret
= create_pending_snapshot(trans
, pending
);
1922 static void update_super_roots(struct btrfs_fs_info
*fs_info
)
1924 struct btrfs_root_item
*root_item
;
1925 struct btrfs_super_block
*super
;
1927 super
= fs_info
->super_copy
;
1929 root_item
= &fs_info
->chunk_root
->root_item
;
1930 super
->chunk_root
= root_item
->bytenr
;
1931 super
->chunk_root_generation
= root_item
->generation
;
1932 super
->chunk_root_level
= root_item
->level
;
1934 root_item
= &fs_info
->tree_root
->root_item
;
1935 super
->root
= root_item
->bytenr
;
1936 super
->generation
= root_item
->generation
;
1937 super
->root_level
= root_item
->level
;
1938 if (btrfs_test_opt(fs_info
, SPACE_CACHE
))
1939 super
->cache_generation
= root_item
->generation
;
1940 else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1
, &fs_info
->flags
))
1941 super
->cache_generation
= 0;
1942 if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN
, &fs_info
->flags
))
1943 super
->uuid_tree_generation
= root_item
->generation
;
1946 int btrfs_transaction_in_commit(struct btrfs_fs_info
*info
)
1948 struct btrfs_transaction
*trans
;
1951 spin_lock(&info
->trans_lock
);
1952 trans
= info
->running_transaction
;
1954 ret
= (trans
->state
>= TRANS_STATE_COMMIT_START
);
1955 spin_unlock(&info
->trans_lock
);
1959 int btrfs_transaction_blocked(struct btrfs_fs_info
*info
)
1961 struct btrfs_transaction
*trans
;
1964 spin_lock(&info
->trans_lock
);
1965 trans
= info
->running_transaction
;
1967 ret
= is_transaction_blocked(trans
);
1968 spin_unlock(&info
->trans_lock
);
1972 void btrfs_commit_transaction_async(struct btrfs_trans_handle
*trans
)
1974 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1975 struct btrfs_transaction
*cur_trans
;
1977 /* Kick the transaction kthread. */
1978 set_bit(BTRFS_FS_COMMIT_TRANS
, &fs_info
->flags
);
1979 wake_up_process(fs_info
->transaction_kthread
);
1981 /* take transaction reference */
1982 cur_trans
= trans
->transaction
;
1983 refcount_inc(&cur_trans
->use_count
);
1985 btrfs_end_transaction(trans
);
1988 * Wait for the current transaction commit to start and block
1989 * subsequent transaction joins
1991 btrfs_might_wait_for_state(fs_info
, BTRFS_LOCKDEP_TRANS_COMMIT_PREP
);
1992 wait_event(fs_info
->transaction_blocked_wait
,
1993 cur_trans
->state
>= TRANS_STATE_COMMIT_START
||
1994 TRANS_ABORTED(cur_trans
));
1995 btrfs_put_transaction(cur_trans
);
1998 static void cleanup_transaction(struct btrfs_trans_handle
*trans
, int err
)
2000 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2001 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
2003 WARN_ON(refcount_read(&trans
->use_count
) > 1);
2005 btrfs_abort_transaction(trans
, err
);
2007 spin_lock(&fs_info
->trans_lock
);
2010 * If the transaction is removed from the list, it means this
2011 * transaction has been committed successfully, so it is impossible
2012 * to call the cleanup function.
2014 BUG_ON(list_empty(&cur_trans
->list
));
2016 if (cur_trans
== fs_info
->running_transaction
) {
2017 cur_trans
->state
= TRANS_STATE_COMMIT_DOING
;
2018 spin_unlock(&fs_info
->trans_lock
);
2021 * The thread has already released the lockdep map as reader
2022 * already in btrfs_commit_transaction().
2024 btrfs_might_wait_for_event(fs_info
, btrfs_trans_num_writers
);
2025 wait_event(cur_trans
->writer_wait
,
2026 atomic_read(&cur_trans
->num_writers
) == 1);
2028 spin_lock(&fs_info
->trans_lock
);
2032 * Now that we know no one else is still using the transaction we can
2033 * remove the transaction from the list of transactions. This avoids
2034 * the transaction kthread from cleaning up the transaction while some
2035 * other task is still using it, which could result in a use-after-free
2036 * on things like log trees, as it forces the transaction kthread to
2037 * wait for this transaction to be cleaned up by us.
2039 list_del_init(&cur_trans
->list
);
2041 spin_unlock(&fs_info
->trans_lock
);
2043 btrfs_cleanup_one_transaction(trans
->transaction
, fs_info
);
2045 spin_lock(&fs_info
->trans_lock
);
2046 if (cur_trans
== fs_info
->running_transaction
)
2047 fs_info
->running_transaction
= NULL
;
2048 spin_unlock(&fs_info
->trans_lock
);
2050 if (trans
->type
& __TRANS_FREEZABLE
)
2051 sb_end_intwrite(fs_info
->sb
);
2052 btrfs_put_transaction(cur_trans
);
2053 btrfs_put_transaction(cur_trans
);
2055 trace_btrfs_transaction_commit(fs_info
);
2057 if (current
->journal_info
== trans
)
2058 current
->journal_info
= NULL
;
2061 * If relocation is running, we can't cancel scrub because that will
2062 * result in a deadlock. Before relocating a block group, relocation
2063 * pauses scrub, then starts and commits a transaction before unpausing
2064 * scrub. If the transaction commit is being done by the relocation
2065 * task or triggered by another task and the relocation task is waiting
2066 * for the commit, and we end up here due to an error in the commit
2067 * path, then calling btrfs_scrub_cancel() will deadlock, as we are
2068 * asking for scrub to stop while having it asked to be paused higher
2069 * above in relocation code.
2071 if (!test_bit(BTRFS_FS_RELOC_RUNNING
, &fs_info
->flags
))
2072 btrfs_scrub_cancel(fs_info
);
2074 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
2078 * Release reserved delayed ref space of all pending block groups of the
2079 * transaction and remove them from the list
2081 static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle
*trans
)
2083 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2084 struct btrfs_block_group
*block_group
, *tmp
;
2086 list_for_each_entry_safe(block_group
, tmp
, &trans
->new_bgs
, bg_list
) {
2087 btrfs_delayed_refs_rsv_release(fs_info
, 1);
2088 list_del_init(&block_group
->bg_list
);
2092 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info
*fs_info
)
2095 * We use try_to_writeback_inodes_sb() here because if we used
2096 * btrfs_start_delalloc_roots we would deadlock with fs freeze.
2097 * Currently are holding the fs freeze lock, if we do an async flush
2098 * we'll do btrfs_join_transaction() and deadlock because we need to
2099 * wait for the fs freeze lock. Using the direct flushing we benefit
2100 * from already being in a transaction and our join_transaction doesn't
2101 * have to re-take the fs freeze lock.
2103 * Note that try_to_writeback_inodes_sb() will only trigger writeback
2104 * if it can read lock sb->s_umount. It will always be able to lock it,
2105 * except when the filesystem is being unmounted or being frozen, but in
2106 * those cases sync_filesystem() is called, which results in calling
2107 * writeback_inodes_sb() while holding a write lock on sb->s_umount.
2108 * Note that we don't call writeback_inodes_sb() directly, because it
2109 * will emit a warning if sb->s_umount is not locked.
2111 if (btrfs_test_opt(fs_info
, FLUSHONCOMMIT
))
2112 try_to_writeback_inodes_sb(fs_info
->sb
, WB_REASON_SYNC
);
2116 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info
*fs_info
)
2118 if (btrfs_test_opt(fs_info
, FLUSHONCOMMIT
))
2119 btrfs_wait_ordered_roots(fs_info
, U64_MAX
, 0, (u64
)-1);
2123 * Add a pending snapshot associated with the given transaction handle to the
2124 * respective handle. This must be called after the transaction commit started
2125 * and while holding fs_info->trans_lock.
2126 * This serves to guarantee a caller of btrfs_commit_transaction() that it can
2127 * safely free the pending snapshot pointer in case btrfs_commit_transaction()
2130 static void add_pending_snapshot(struct btrfs_trans_handle
*trans
)
2132 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
2134 if (!trans
->pending_snapshot
)
2137 lockdep_assert_held(&trans
->fs_info
->trans_lock
);
2138 ASSERT(cur_trans
->state
>= TRANS_STATE_COMMIT_PREP
);
2140 list_add(&trans
->pending_snapshot
->list
, &cur_trans
->pending_snapshots
);
2143 static void update_commit_stats(struct btrfs_fs_info
*fs_info
, ktime_t interval
)
2145 fs_info
->commit_stats
.commit_count
++;
2146 fs_info
->commit_stats
.last_commit_dur
= interval
;
2147 fs_info
->commit_stats
.max_commit_dur
=
2148 max_t(u64
, fs_info
->commit_stats
.max_commit_dur
, interval
);
2149 fs_info
->commit_stats
.total_commit_dur
+= interval
;
2152 int btrfs_commit_transaction(struct btrfs_trans_handle
*trans
)
2154 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2155 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
2156 struct btrfs_transaction
*prev_trans
= NULL
;
2161 ASSERT(refcount_read(&trans
->use_count
) == 1);
2162 btrfs_trans_state_lockdep_acquire(fs_info
, BTRFS_LOCKDEP_TRANS_COMMIT_PREP
);
2164 clear_bit(BTRFS_FS_NEED_TRANS_COMMIT
, &fs_info
->flags
);
2166 /* Stop the commit early if ->aborted is set */
2167 if (TRANS_ABORTED(cur_trans
)) {
2168 ret
= cur_trans
->aborted
;
2169 goto lockdep_trans_commit_start_release
;
2172 btrfs_trans_release_metadata(trans
);
2173 trans
->block_rsv
= NULL
;
2176 * We only want one transaction commit doing the flushing so we do not
2177 * waste a bunch of time on lock contention on the extent root node.
2179 if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING
,
2180 &cur_trans
->delayed_refs
.flags
)) {
2182 * Make a pass through all the delayed refs we have so far.
2183 * Any running threads may add more while we are here.
2185 ret
= btrfs_run_delayed_refs(trans
, 0);
2187 goto lockdep_trans_commit_start_release
;
2190 btrfs_create_pending_block_groups(trans
);
2192 if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN
, &cur_trans
->flags
)) {
2195 /* this mutex is also taken before trying to set
2196 * block groups readonly. We need to make sure
2197 * that nobody has set a block group readonly
2198 * after a extents from that block group have been
2199 * allocated for cache files. btrfs_set_block_group_ro
2200 * will wait for the transaction to commit if it
2201 * finds BTRFS_TRANS_DIRTY_BG_RUN set.
2203 * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure
2204 * only one process starts all the block group IO. It wouldn't
2205 * hurt to have more than one go through, but there's no
2206 * real advantage to it either.
2208 mutex_lock(&fs_info
->ro_block_group_mutex
);
2209 if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN
,
2212 mutex_unlock(&fs_info
->ro_block_group_mutex
);
2215 ret
= btrfs_start_dirty_block_groups(trans
);
2217 goto lockdep_trans_commit_start_release
;
2221 spin_lock(&fs_info
->trans_lock
);
2222 if (cur_trans
->state
>= TRANS_STATE_COMMIT_PREP
) {
2223 enum btrfs_trans_state want_state
= TRANS_STATE_COMPLETED
;
2225 add_pending_snapshot(trans
);
2227 spin_unlock(&fs_info
->trans_lock
);
2228 refcount_inc(&cur_trans
->use_count
);
2230 if (trans
->in_fsync
)
2231 want_state
= TRANS_STATE_SUPER_COMMITTED
;
2233 btrfs_trans_state_lockdep_release(fs_info
,
2234 BTRFS_LOCKDEP_TRANS_COMMIT_PREP
);
2235 ret
= btrfs_end_transaction(trans
);
2236 wait_for_commit(cur_trans
, want_state
);
2238 if (TRANS_ABORTED(cur_trans
))
2239 ret
= cur_trans
->aborted
;
2241 btrfs_put_transaction(cur_trans
);
2246 cur_trans
->state
= TRANS_STATE_COMMIT_PREP
;
2247 wake_up(&fs_info
->transaction_blocked_wait
);
2248 btrfs_trans_state_lockdep_release(fs_info
, BTRFS_LOCKDEP_TRANS_COMMIT_PREP
);
2250 if (cur_trans
->list
.prev
!= &fs_info
->trans_list
) {
2251 enum btrfs_trans_state want_state
= TRANS_STATE_COMPLETED
;
2253 if (trans
->in_fsync
)
2254 want_state
= TRANS_STATE_SUPER_COMMITTED
;
2256 prev_trans
= list_entry(cur_trans
->list
.prev
,
2257 struct btrfs_transaction
, list
);
2258 if (prev_trans
->state
< want_state
) {
2259 refcount_inc(&prev_trans
->use_count
);
2260 spin_unlock(&fs_info
->trans_lock
);
2262 wait_for_commit(prev_trans
, want_state
);
2264 ret
= READ_ONCE(prev_trans
->aborted
);
2266 btrfs_put_transaction(prev_trans
);
2268 goto lockdep_release
;
2269 spin_lock(&fs_info
->trans_lock
);
2273 * The previous transaction was aborted and was already removed
2274 * from the list of transactions at fs_info->trans_list. So we
2275 * abort to prevent writing a new superblock that reflects a
2276 * corrupt state (pointing to trees with unwritten nodes/leafs).
2278 if (BTRFS_FS_ERROR(fs_info
)) {
2279 spin_unlock(&fs_info
->trans_lock
);
2281 goto lockdep_release
;
2285 cur_trans
->state
= TRANS_STATE_COMMIT_START
;
2286 wake_up(&fs_info
->transaction_blocked_wait
);
2287 spin_unlock(&fs_info
->trans_lock
);
2290 * Get the time spent on the work done by the commit thread and not
2291 * the time spent waiting on a previous commit
2293 start_time
= ktime_get_ns();
2295 extwriter_counter_dec(cur_trans
, trans
->type
);
2297 ret
= btrfs_start_delalloc_flush(fs_info
);
2299 goto lockdep_release
;
2301 ret
= btrfs_run_delayed_items(trans
);
2303 goto lockdep_release
;
2306 * The thread has started/joined the transaction thus it holds the
2307 * lockdep map as a reader. It has to release it before acquiring the
2308 * lockdep map as a writer.
2310 btrfs_lockdep_release(fs_info
, btrfs_trans_num_extwriters
);
2311 btrfs_might_wait_for_event(fs_info
, btrfs_trans_num_extwriters
);
2312 wait_event(cur_trans
->writer_wait
,
2313 extwriter_counter_read(cur_trans
) == 0);
2315 /* some pending stuffs might be added after the previous flush. */
2316 ret
= btrfs_run_delayed_items(trans
);
2318 btrfs_lockdep_release(fs_info
, btrfs_trans_num_writers
);
2319 goto cleanup_transaction
;
2322 btrfs_wait_delalloc_flush(fs_info
);
2325 * Wait for all ordered extents started by a fast fsync that joined this
2326 * transaction. Otherwise if this transaction commits before the ordered
2327 * extents complete we lose logged data after a power failure.
2329 btrfs_might_wait_for_event(fs_info
, btrfs_trans_pending_ordered
);
2330 wait_event(cur_trans
->pending_wait
,
2331 atomic_read(&cur_trans
->pending_ordered
) == 0);
2333 btrfs_scrub_pause(fs_info
);
2335 * Ok now we need to make sure to block out any other joins while we
2336 * commit the transaction. We could have started a join before setting
2337 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
2339 spin_lock(&fs_info
->trans_lock
);
2340 add_pending_snapshot(trans
);
2341 cur_trans
->state
= TRANS_STATE_COMMIT_DOING
;
2342 spin_unlock(&fs_info
->trans_lock
);
2345 * The thread has started/joined the transaction thus it holds the
2346 * lockdep map as a reader. It has to release it before acquiring the
2347 * lockdep map as a writer.
2349 btrfs_lockdep_release(fs_info
, btrfs_trans_num_writers
);
2350 btrfs_might_wait_for_event(fs_info
, btrfs_trans_num_writers
);
2351 wait_event(cur_trans
->writer_wait
,
2352 atomic_read(&cur_trans
->num_writers
) == 1);
2355 * Make lockdep happy by acquiring the state locks after
2356 * btrfs_trans_num_writers is released. If we acquired the state locks
2357 * before releasing the btrfs_trans_num_writers lock then lockdep would
2358 * complain because we did not follow the reverse order unlocking rule.
2360 btrfs_trans_state_lockdep_acquire(fs_info
, BTRFS_LOCKDEP_TRANS_COMPLETED
);
2361 btrfs_trans_state_lockdep_acquire(fs_info
, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED
);
2362 btrfs_trans_state_lockdep_acquire(fs_info
, BTRFS_LOCKDEP_TRANS_UNBLOCKED
);
2365 * We've started the commit, clear the flag in case we were triggered to
2366 * do an async commit but somebody else started before the transaction
2367 * kthread could do the work.
2369 clear_bit(BTRFS_FS_COMMIT_TRANS
, &fs_info
->flags
);
2371 if (TRANS_ABORTED(cur_trans
)) {
2372 ret
= cur_trans
->aborted
;
2373 btrfs_trans_state_lockdep_release(fs_info
, BTRFS_LOCKDEP_TRANS_UNBLOCKED
);
2374 goto scrub_continue
;
2377 * the reloc mutex makes sure that we stop
2378 * the balancing code from coming in and moving
2379 * extents around in the middle of the commit
2381 mutex_lock(&fs_info
->reloc_mutex
);
2384 * We needn't worry about the delayed items because we will
2385 * deal with them in create_pending_snapshot(), which is the
2386 * core function of the snapshot creation.
2388 ret
= create_pending_snapshots(trans
);
2393 * We insert the dir indexes of the snapshots and update the inode
2394 * of the snapshots' parents after the snapshot creation, so there
2395 * are some delayed items which are not dealt with. Now deal with
2398 * We needn't worry that this operation will corrupt the snapshots,
2399 * because all the tree which are snapshoted will be forced to COW
2400 * the nodes and leaves.
2402 ret
= btrfs_run_delayed_items(trans
);
2406 ret
= btrfs_run_delayed_refs(trans
, (unsigned long)-1);
2411 * make sure none of the code above managed to slip in a
2414 btrfs_assert_delayed_root_empty(fs_info
);
2416 WARN_ON(cur_trans
!= trans
->transaction
);
2418 ret
= commit_fs_roots(trans
);
2422 /* commit_fs_roots gets rid of all the tree log roots, it is now
2423 * safe to free the root of tree log roots
2425 btrfs_free_log_root_tree(trans
, fs_info
);
2428 * Since fs roots are all committed, we can get a quite accurate
2429 * new_roots. So let's do quota accounting.
2431 ret
= btrfs_qgroup_account_extents(trans
);
2435 ret
= commit_cowonly_roots(trans
);
2440 * The tasks which save the space cache and inode cache may also
2441 * update ->aborted, check it.
2443 if (TRANS_ABORTED(cur_trans
)) {
2444 ret
= cur_trans
->aborted
;
2448 cur_trans
= fs_info
->running_transaction
;
2450 btrfs_set_root_node(&fs_info
->tree_root
->root_item
,
2451 fs_info
->tree_root
->node
);
2452 list_add_tail(&fs_info
->tree_root
->dirty_list
,
2453 &cur_trans
->switch_commits
);
2455 btrfs_set_root_node(&fs_info
->chunk_root
->root_item
,
2456 fs_info
->chunk_root
->node
);
2457 list_add_tail(&fs_info
->chunk_root
->dirty_list
,
2458 &cur_trans
->switch_commits
);
2460 if (btrfs_fs_incompat(fs_info
, EXTENT_TREE_V2
)) {
2461 btrfs_set_root_node(&fs_info
->block_group_root
->root_item
,
2462 fs_info
->block_group_root
->node
);
2463 list_add_tail(&fs_info
->block_group_root
->dirty_list
,
2464 &cur_trans
->switch_commits
);
2467 switch_commit_roots(trans
);
2469 ASSERT(list_empty(&cur_trans
->dirty_bgs
));
2470 ASSERT(list_empty(&cur_trans
->io_bgs
));
2471 update_super_roots(fs_info
);
2473 btrfs_set_super_log_root(fs_info
->super_copy
, 0);
2474 btrfs_set_super_log_root_level(fs_info
->super_copy
, 0);
2475 memcpy(fs_info
->super_for_commit
, fs_info
->super_copy
,
2476 sizeof(*fs_info
->super_copy
));
2478 btrfs_commit_device_sizes(cur_trans
);
2480 clear_bit(BTRFS_FS_LOG1_ERR
, &fs_info
->flags
);
2481 clear_bit(BTRFS_FS_LOG2_ERR
, &fs_info
->flags
);
2483 btrfs_trans_release_chunk_metadata(trans
);
2486 * Before changing the transaction state to TRANS_STATE_UNBLOCKED and
2487 * setting fs_info->running_transaction to NULL, lock tree_log_mutex to
2488 * make sure that before we commit our superblock, no other task can
2489 * start a new transaction and commit a log tree before we commit our
2490 * superblock. Anyone trying to commit a log tree locks this mutex before
2491 * writing its superblock.
2493 mutex_lock(&fs_info
->tree_log_mutex
);
2495 spin_lock(&fs_info
->trans_lock
);
2496 cur_trans
->state
= TRANS_STATE_UNBLOCKED
;
2497 fs_info
->running_transaction
= NULL
;
2498 spin_unlock(&fs_info
->trans_lock
);
2499 mutex_unlock(&fs_info
->reloc_mutex
);
2501 wake_up(&fs_info
->transaction_wait
);
2502 btrfs_trans_state_lockdep_release(fs_info
, BTRFS_LOCKDEP_TRANS_UNBLOCKED
);
2504 /* If we have features changed, wake up the cleaner to update sysfs. */
2505 if (test_bit(BTRFS_FS_FEATURE_CHANGED
, &fs_info
->flags
) &&
2506 fs_info
->cleaner_kthread
)
2507 wake_up_process(fs_info
->cleaner_kthread
);
2509 ret
= btrfs_write_and_wait_transaction(trans
);
2511 btrfs_handle_fs_error(fs_info
, ret
,
2512 "Error while writing out transaction");
2513 mutex_unlock(&fs_info
->tree_log_mutex
);
2514 goto scrub_continue
;
2517 ret
= write_all_supers(fs_info
, 0);
2519 * the super is written, we can safely allow the tree-loggers
2520 * to go about their business
2522 mutex_unlock(&fs_info
->tree_log_mutex
);
2524 goto scrub_continue
;
2527 * We needn't acquire the lock here because there is no other task
2528 * which can change it.
2530 cur_trans
->state
= TRANS_STATE_SUPER_COMMITTED
;
2531 wake_up(&cur_trans
->commit_wait
);
2532 btrfs_trans_state_lockdep_release(fs_info
, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED
);
2534 btrfs_finish_extent_commit(trans
);
2536 if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS
, &cur_trans
->flags
))
2537 btrfs_clear_space_info_full(fs_info
);
2539 fs_info
->last_trans_committed
= cur_trans
->transid
;
2541 * We needn't acquire the lock here because there is no other task
2542 * which can change it.
2544 cur_trans
->state
= TRANS_STATE_COMPLETED
;
2545 wake_up(&cur_trans
->commit_wait
);
2546 btrfs_trans_state_lockdep_release(fs_info
, BTRFS_LOCKDEP_TRANS_COMPLETED
);
2548 spin_lock(&fs_info
->trans_lock
);
2549 list_del_init(&cur_trans
->list
);
2550 spin_unlock(&fs_info
->trans_lock
);
2552 btrfs_put_transaction(cur_trans
);
2553 btrfs_put_transaction(cur_trans
);
2555 if (trans
->type
& __TRANS_FREEZABLE
)
2556 sb_end_intwrite(fs_info
->sb
);
2558 trace_btrfs_transaction_commit(fs_info
);
2560 interval
= ktime_get_ns() - start_time
;
2562 btrfs_scrub_continue(fs_info
);
2564 if (current
->journal_info
== trans
)
2565 current
->journal_info
= NULL
;
2567 kmem_cache_free(btrfs_trans_handle_cachep
, trans
);
2569 update_commit_stats(fs_info
, interval
);
2574 mutex_unlock(&fs_info
->reloc_mutex
);
2575 btrfs_trans_state_lockdep_release(fs_info
, BTRFS_LOCKDEP_TRANS_UNBLOCKED
);
2577 btrfs_trans_state_lockdep_release(fs_info
, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED
);
2578 btrfs_trans_state_lockdep_release(fs_info
, BTRFS_LOCKDEP_TRANS_COMPLETED
);
2579 btrfs_scrub_continue(fs_info
);
2580 cleanup_transaction
:
2581 btrfs_trans_release_metadata(trans
);
2582 btrfs_cleanup_pending_block_groups(trans
);
2583 btrfs_trans_release_chunk_metadata(trans
);
2584 trans
->block_rsv
= NULL
;
2585 btrfs_warn(fs_info
, "Skipping commit of aborted transaction.");
2586 if (current
->journal_info
== trans
)
2587 current
->journal_info
= NULL
;
2588 cleanup_transaction(trans
, ret
);
2593 btrfs_lockdep_release(fs_info
, btrfs_trans_num_extwriters
);
2594 btrfs_lockdep_release(fs_info
, btrfs_trans_num_writers
);
2595 goto cleanup_transaction
;
2597 lockdep_trans_commit_start_release
:
2598 btrfs_trans_state_lockdep_release(fs_info
, BTRFS_LOCKDEP_TRANS_COMMIT_PREP
);
2599 btrfs_end_transaction(trans
);
2604 * return < 0 if error
2605 * 0 if there are no more dead_roots at the time of call
2606 * 1 there are more to be processed, call me again
2608 * The return value indicates there are certainly more snapshots to delete, but
2609 * if there comes a new one during processing, it may return 0. We don't mind,
2610 * because btrfs_commit_super will poke cleaner thread and it will process it a
2611 * few seconds later.
2613 int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info
*fs_info
)
2615 struct btrfs_root
*root
;
2618 spin_lock(&fs_info
->trans_lock
);
2619 if (list_empty(&fs_info
->dead_roots
)) {
2620 spin_unlock(&fs_info
->trans_lock
);
2623 root
= list_first_entry(&fs_info
->dead_roots
,
2624 struct btrfs_root
, root_list
);
2625 list_del_init(&root
->root_list
);
2626 spin_unlock(&fs_info
->trans_lock
);
2628 btrfs_debug(fs_info
, "cleaner removing %llu", root
->root_key
.objectid
);
2630 btrfs_kill_all_delayed_nodes(root
);
2632 if (btrfs_header_backref_rev(root
->node
) <
2633 BTRFS_MIXED_BACKREF_REV
)
2634 ret
= btrfs_drop_snapshot(root
, 0, 0);
2636 ret
= btrfs_drop_snapshot(root
, 1, 0);
2638 btrfs_put_root(root
);
2639 return (ret
< 0) ? 0 : 1;
2643 * We only mark the transaction aborted and then set the file system read-only.
2644 * This will prevent new transactions from starting or trying to join this
2647 * This means that error recovery at the call site is limited to freeing
2648 * any local memory allocations and passing the error code up without
2649 * further cleanup. The transaction should complete as it normally would
2650 * in the call path but will return -EIO.
2652 * We'll complete the cleanup in btrfs_end_transaction and
2653 * btrfs_commit_transaction.
2655 void __cold
__btrfs_abort_transaction(struct btrfs_trans_handle
*trans
,
2656 const char *function
,
2657 unsigned int line
, int errno
, bool first_hit
)
2659 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2661 WRITE_ONCE(trans
->aborted
, errno
);
2662 WRITE_ONCE(trans
->transaction
->aborted
, errno
);
2663 if (first_hit
&& errno
== -ENOSPC
)
2664 btrfs_dump_space_info_for_trans_abort(fs_info
);
2665 /* Wake up anybody who may be waiting on this transaction */
2666 wake_up(&fs_info
->transaction_wait
);
2667 wake_up(&fs_info
->transaction_blocked_wait
);
2668 __btrfs_handle_fs_error(fs_info
, function
, line
, errno
, NULL
);
2671 int __init
btrfs_transaction_init(void)
2673 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle",
2674 sizeof(struct btrfs_trans_handle
), 0,
2675 SLAB_TEMPORARY
| SLAB_MEM_SPREAD
, NULL
);
2676 if (!btrfs_trans_handle_cachep
)
2681 void __cold
btrfs_transaction_exit(void)
2683 kmem_cache_destroy(btrfs_trans_handle_cachep
);